Conference Papers - WE Upjohn Institute

Transcription

900 7th Street, NW
Suite 600
Washington, D.C. 20001
Phone: (202) 347-3190
Fax: (202) 393-0993
www.napawash.org
MEASUREMENT ISSUES ARISING FROM THE GROWTH OF GLOBALIZATION
CONFERENCE PAPERS—NOVEMBER 6-7, 2009, WASHINGTON DC
300 S. Westnedge Avenue
Kalamazoo, Michigan 49007-4686
Telephone: 269/343-5541
Fax: 269/343-3308
www.upjohninst.org/
MEASUREMENT ISSUES
ARISING FROM THE
GROWTH OF GLOBALIZATION
CONFERENCE PAPERS
November 6-7, 2009, Washington DC
W.E. Upjohn Institute for Employment Research
National Academy of Public Administration
August 2010
The conference was funded with grants from
the Alfred P. Sloan Foundation and the U. S. Bureau of Economic Analysis
About the Confe re nce Or gani zer s:
The National Academy of Public Administration (NAPA) is a non-profit,
independent organization of top public management and organizational leaders
who tackle the nation’s most critical and complex public management
challenges. With a network of more than 650 distinguished Fellows and an
experienced professional staff, the National Academy is uniquely qualifi ed and
trusted across government to provide objective advice and practical solutions
based on systematic research and expert analysis. Established in 1967 and
chartered by Congress in 1984, the National Academy continues to make a
positive impact by helping federal, state and local governments respond
effectively to current circumstances and changing conditions. Learn more about
the National Academy and its work at www.NAPAwash.org
The W.E. Upjohn Institute for Employment Research is a nonprofit research
organization devoted to finding and promoting solutions to employment-related
problems at the national, state, and local levels. It is an activity of the W.E. Upjohn
Unemployment Trustee Corporation, which was established in 1932 to administer
a fund set aside by Dr. W.E. Upjohn, founder of The Upjohn Company, to seek
ways to counteract the loss of employment income during economic downturns.
Today, the broad objectives of the Institute’s research, grant, and publication
programs are to 1) promote scholarship and experimentation on issues of public
and private employment and unemployment policy, and 2) make knowledge and
scholarship relevant and useful to policymakers in their pursuit of solutions to
employment and unemployment problems.
MEASUREMENT ISSUES
ARISING FROM THE
CONFERENCE PAPERS
November 6-7, 2009
Washington, DC 20001
Editors
Susan N. Houseman, Senior Economist, W.E. Upjohn Institute for Employment Research
Kenneth F. Ryder, Jr., Fellow, National Academy for Public Administration
Officers of the Academy
Kenneth S. Apfel,* Chair of the Board
Timothy C. Clark,* Vice Chair
*
Jennifer L. Dorn, President and Chief Executive Officer
Diane M. Disney,* Secretary
John J. Callahan,* Treasurer
Planning Group
Susan N. Houseman, W.E. Upjohn Institute for Employment Research
Kenneth F. Ryder Jr.,* National Academy of Public Administration
Katharine Abraham, University of Maryland
William F. Alterman, Bureau of Labor Statistics
Michael J. Harper, Bureau of Labor Statistics
Michael W. Horrigan, Bureau of Labor Statistics
Thomas L. Mesenbourg, Jr., Bureau of the Census
Brent R. Moulton, Bureau of Economic Analysis
Alice Nakamura, University of Alberta, Canada
Janet L. Norwood,* National Academy of Public Administration
Timothy J. Sturgeon, Massachusetts Institute of Technology
Obie G. Whichard, Bureau of Economic Analysis
Robert E. Yuskavage, Bureau of Economic Analysis
Staff
Lena E. Trudeau, Vice President
Susan N. Houseman, Co-Project Director+
Kenneth F. Ryder, Jr.,* Co-Project Director
Lillian Vesic-Petrovic, Senior Research Analyst+
Sue A. Berkebile, Administrative Assistant+
Morgan Clark, Research Associate
Martha S. Ditmeyer, Senior Program Associate
*
National Academy Fellow
+
Staff, W.E. Upjohn Institute for Employment Research
____________________________________________________________________________________________________________________
The views expressed in this report are those of the Conference participants. They do not
necessarily reflect the views of W.E. Upjohn Institute and the National Academy as institutions.
900 7th Street, N.W.
Suite 600
Washington, DC 20001-3888
www.napawash.org
Published August 2010
Printed in the United States of America
Academy Project Number: 2131-000
ii
CONFERENCE AGENDA
MEASUREMENT ISSUES ARISING
FROM THE GROWTH OF GLOBALIZATION∗
November 6-7, 2009
Washington, DC
and the
900 7th Street NW, Suite 600
Washington, DC
(202) 347-3190
FRIDAY NOVEMBER 6
SESSION 1: SERVICES OFFSHORING
Chair: Kenneth Ryder (NAPA)
Measuring the Impact of Trade in Services: Prospects and Challenges, J. Bradford Jensen
(Georgetown University and Peterson Institute for International Economics)
Measuring Success in the Global Economy: International Trade, Industrial Upgrading, and
Business Function Outsourcing in Global Value Chains, Timothy Sturgeon (Massachusetts
Institute of Technology) and Gary Gereffi (Duke University)
Discussant: Torbjorn Fredriksson (UNCTAD)
∗
The co nference w as funded w ith grants f rom th e A lfred P. Slo an Foundation and th e Bur eau of Econ omic
Analysis.
iii
SESSION 2: MEASURING THE IMPACT OF OFFSHORING ON THE LABOR
MARKET
Chair: Janet Norwood (NAPA)
Addressing the Demand for Time Series and Longitudinal Data on Occupational
Employment, Katharine Abraham (University of Maryland) and James R. Spletzer (Bureau of
Labor Statistics)
Understanding the Domestic Labor Market Impact of Offshore Services Outsourcing:
Measurement Issues, Lori Kletzer (University of California-Santa Cruz and Peterson Institute
for International Economics)
Discussant: Timothy Sturgeon (MIT)
SESSION 3: MEASUREMENT IMPLICATIONS OF TRANSFORMATION
OFFSHORING AND IMPORTED INTERMEDIATE INPUTS
Chair: Ned Howenstein (Bureau of Economic Analysis)
Outsourcing, Offshoring, and Trade: Identifying Foreign Activity across Census Data
Products, Ron Jarmin, C. Krizan, and John Tang (U.S. Census Bureau)
Effects of Imported Intermediate Inputs on Productivity. Lucy P. Eldridge and Michael J.
Harper (Bureau of Labor Statistics)
Discussant: John Haltiwanger
SESSION 4: OFFSHORING AND PRICE MEASUREMENT: INDUSTRY CASE
STUDIES
Chair: Jon Steinsson (Columbia University)
Offshoring and Price Measurement in the Semiconductor Industry, David Byrne (Federal
Reserve Board), Brian Kovak (University of Michigan), and Ryan Michaels (University of
Michigan)
Globalization, Prices of IT Software & Services Measurement Issues, Catherine Mann
(Brandeis University and Peterson Institute for International Economics)
Imports of Intermediate Parts in the Auto Industry—A Case Study, Thomas H. Klier (Federal
Reserve Bank of Chicago) and James M. Rubenstein (University of Miami, Ohio)
Discussant: Kimberly Zieschang (IMF)
iv
SATURDAY NOVEMBER 7
SESSION 5: MEASUREMENT OF IMPORT PRICES
Chair: Michael Horrigan (Bureau of Labor Statistics)
Bias Due to Input Source Substitutions: Can It Be Measured? Erwin Diewert (University of
British Columbia) and Alice Nakamura (University of Alberta)
Are there Unmeasured Declines in Prices of Imported Final Consumption Goods?, Marshall
Reinsdorf and Robert Yuskavage (Bureau of Economic Analysis)
Discussant: Barry Bosworth (Brookings Institution)
SESSION 6: MEASUREMENT OF IMPORT AND INPUT PRICES
Chair: Julia Lane (National Science Foundation)
Offshoring and the State of American Manufacturing Susan Houseman (W.E. Upjohn
Institute), Christopher Kurz (Federal Reserve Board), Paul Lengermann (Federal Reserve
Board), and Benjamin Mandel (Federal Reserve Board)
Producing an Input Price Index, William Alterman (Bureau of Labor Statistics)
Discussant: Emi Nakamura (Columbia University)
SESSION 7: MEASUREMENT OF IMPORT PRICES AND IMPORT USES:
EVIDENCE OF BIASES
Chair: Carol Corrado (Conference Board)
Imported Inputs and Industry Contributions to Economic Growth: An Assessment of
Alternative Approaches, Erich Strassner, Robert Yuskavage and Jennifer Lee (Bureau of
Economic Analysis)
Evaluating Estimates of Materials Offshoring from U.S. Manufacturing, Robert C. Feenstra
(University of California-Davis) and J. Bradford Jensen (Georgetown University and
Peterson Institute for International Economics)
Discussant: William Milberg (New School for Social Research)
Errors from the Proportionality Assumption in the Measurement of Offshoring:
Application to German Labor Demand, Deborah Winkler and William Milberg (Schwartz
Center for Economic Policy Analysis, The New School for Social Research)
v
PANEL DISCUSSION: PRIORITIZING STEPS FOR THE STATISTICAL AGENCIES
Chair: Susan Houseman (W.E. Upjohn Institute)
Panel members: Katharine Abraham (University of Maryland), Erwin Diewert (University of
British Columbia), Robert C. Feenstra (University of California-Davis), Keith Hall (Bureau
of Labor Statistics), and Steven Landefeld (Bureau of Economic Analysis)
CONCLUDING COMMENTS AND ADJOURNMENT
Susan Houseman and Kenneth Ryder
vi
CONFERENCE ATTENDEES
Government Accountability Office
Giulia Cangiano
Loren Yager
Bureau of Economic Analysis
Ana Aizcorbe
Maria Borga
Ned Howenstein
Richard Kane
J. Steven Landefeld
Jennifer Lee
Marshall Reinsdorf
Erich Strassner
Obie Whichard
Robert Yuskavage
William Zeile
Federal Reserve Bank of Chicago
Thomas Klier
Federal Reserve Board
David Byrne
Christopher Kurz
Paul Lengermann
Benjamin Mandel
Jon Stevens
Bureau of Labor Statistics
William Alterman
Anthony Barkume
Sharon Brown
Maureen Doherty
Lucy P. Eldridge
Susan Fleck
Keith Hall
Michael J. Harper
Rodger Hippen
Michael Horrigan
Helen McCulley
Ann Polivka
Connie Sorrentino
James R. Spletzer
Jay Stewart
Leo Sveikauskas
Jim Thomas
Lisa Usher
Jon Weinhagen
Cindy Zoghi
International Monetary Fund
Ralph Kozlow
Kimberly Zieschang
National Academy of Public
Administration
Morgan Clark
Janet Norwood∗
Kenneth Ryder*
National Science Foundation
Julia Lane
UNCTAD
Torbjorn Fredricksson
Universities
Alice Nakamura, University of Alberta,
Canada
Erwin Diewert, University of British
Columbia
Lori Kletzer, University of California, Santa
Cruz and Peterson Institute of
International Economics
Robert C. Feenstra, University of CaliforniaDavis
Emi Nakamura, Columbia University
Census Bureau
C. Krizan
John Tang
Ron Jarmin
∗
vii
National Academy Fellow
Brian Kovak, University of Michigan
Ryan Michaels, University of Michigan
Universities (continued)
Jon Steinsson, Columbia University
J. Bradford Jensen, Georgetown University
Daniel Reyes, Georgetown University
William Milberg , New School for Social
Research
Katharine Abraham, University of Maryland
John Haltiwanger, University of Maryland
Chuck Hulten, University of Maryland
Catherine Mann, Brandeis University and
Peterson Institute for International
Economics
James M. Rubenstein, University of Miami,
Ohio
Timothy Sturgeon, Massachusetts Institute
of Technology
W.E. Upjohn Institute for Employment
Research
Susan N. Houseman
Other Organizations
Barry Bosworth, Brookings Institution
Carol Corrado, Conference Board
Robert Drago, Senate Joint Economic
Committee
Michael Mandel, Business Week
Michael Teitlebaum., Alfred P. Sloan
Foundation
Lou Uchitelle, New York Times
viii
LIST OF CONFERENCE PAPERS
(in order presented)
CONFERENCE ATTENDEES ................................................................................................. vii
SESSION 1: SERVICES OFFSHORING ................................................................................. 1
Measuring the Impact of Trade in Services: Prospects and Challenges (Jensen).................... 3
Measuring Success in the Global Economy: International Trade, Industrial Upgrading,
and Business Function Outsourcing in Global Value Chains (Sturgeon and Gereffi)........... 33
SESSION 2: MEASURING THE IMPACT OF OFFSHORING ON THE LABOR
MARKET............................................................................................................................... 67
Addressing the Demand for Time Series and Longitudinal Data on Occupational
Employment (Abraham and Spletzer) ..................................................................................... 69
Measurement Issues (Kletzer)................................................................................................. 89
SESSION 3: MEASUREMENT IMPLICATIONS OF TRANSFORMATION
OFFSHORING AND IMPORTED INTERMEDIATE INPUTS ........................................ 119
Outsourcing, Offshoring, and Trade: Identifying Foreign Activity across Census Data
Products (Jarmin, Krizan, and Tang).................................................................................... 121
Effects of Imported Intermediate Inputs on Productivity (Eldridge and Harper) ................. 141
SESSION 4: OFFSHORING AND PRICE MEASUREMENT: INDUSTRY CASE
STUDIES ................................................................................................................................... 167
Offshoring and Price Measurement in the Semiconductor Industry (Byrne, Kovak
and Michaels)........................................................................................................................ 169
Globalization, Prices of IT Software & Services: Measurement Issues (Mann) .................. 197
Imports of Intermediate Parts in the Auto Industry—A Case Study
(Klier and Rubenstein).......................................................................................................... 217
ix
SESSION 5: MEASUREMENT OF IMPORT PRICES...................................................... 235
Bias Due to Input Source Substitutions: Can It Be Measured? (Diewert and Nakamura)... 237
Are there Unmeasured Declines in Prices of Imported Final Consumption Goods?
(Reinsdorf and Yuskavage) .................................................................................................. 267
SESSION 6: MEASUREMENT OF IMPORT AND INPUT PRICES............................... 285
Offshoring and the State of American Manufacturing (Houseman, Kurz, Lengermann
and Mandel) .......................................................................................................................... 287
Producing an Input Price Index (Alterman)......................................................................... 315
SESSION 7: MEASUREMENT OF IMPORT PRICES AND IMPORT USES:
EVIDENCE OF BIASES.......................................................................................................... 339
Imported Inputs and Industry Contributions to Economic Growth: An Assessment of
Alternative Approaches (Strassner, Yuskavage and Lee)..................................................... 341
Evaluating Estimates of Materials Offshoring from U.S. Manufacturing
(Feenstra and Jensen)............................................................................................................ 359
Errors from the “Proportionality Assumption” in the Measurement of Offshoring:
Application to German Labor Demand (Winkler and Milberg)........................................... 379
x
SESSION 1: SERVICES OFFSHORING
CHAIR: Kenneth Ryder (NAPA)
Measuring the Impact of Trade in Services: Prospects and Challenges, J. Bradford Jensen
(Georgetown University and Peterson Institute for International Economics)
Measuring Success in the Global Economy: International Trade, Industrial Upgrading, and
Business Function Outsourcing in Global Value Chains, Timothy Sturgeon (Massachusetts
Institute of Technology) and Gary Gereffi (Duke University)
DISCUSSANT: Torbjorn Fredriksson (UNCTAD)
1
Measuring the Impact of Trade in Services:
Prospects and Challenges
J. Bradford Jensen
McDonough School of Business, Georgetown University
Peterson Institute for International Economics
Georgetown Center for Business and Public Policy
October 2009
(revised May 2010)
This paper was prepared for the conference “Measurement Issues Arising from the Growth of
Globalization,” held in Washington, DC, November 6–7, 2009.
3
EXECUTIVE SUMMARY
The large share of employment in the service sector and growing services trade present the
potential for trade in services to have a significant impact on the U.S. economy and highlight the
importance of being able to analyze the impact.
International trade theory and previous empirical work on the manufacturing sector stress several
key considerations for understanding the impact of globalization:
1) The prevalence (how many activities?), scale (how much is being traded?), and direction
(who is trading with whom?)of trade in services.
2) How trade in services evolves over time.
3) The factor intensities used in services provision.
4) The factor intensities across locations.
5) Firm-level heterogeneity (in size, factor intensities, productivity, trade activity) within
and across industries and countries.
Currently available data on the service sector do not support these data needs. Two broad areas
require improvement:
1) Increased industry and geographic detail in trade in services statistics. Current trade in
services statistics are not detailed enough to support robust empirical analysis. Increasing
the detail will require increased resources to collect information from larger sample of
firms, improved access to an adequate sampling frame to support representative
sampling, and lower reporting thresholds.
2) More detailed information on inputs used services production in the United States.
Current data on service sector production within the United States do not provide enough
information on the factor inputs used in production. More information should be
collected on skill intensity, capital intensity, and purchased services. These data should
be collected at the establishment level to the extent possible to increase the industry and
geographic detail available.
Improving our ability to analyze the impact of trade in services will require
1) more funding for service sector data collection, and an
2) improved sampling frame for the Bureau of Economic Analysis’ data collection.
The need for an improved sampling frame and potential efficiencies in data collection suggest
the costs and benefits of moving data collection activities currently performed by the Bureau of
Economic Analysis to the Census Bureau should be investigated.
5
MEASURING THE IMPACT OF TRADE IN SERVICES:
PROSPECTS AND CHALLENGES
The service sector accounts for a large share of employment in the United States. Trade in
services is growing rapidly. The large share of employment in the service sector and growing
services trade present the potential for trade in services to have a significant impact on U.S. firms
and workers.
Despite the potential importance of trade in services, the amount of empirical research on the
impact of trade in services relative to empirical research of the impact of trade in goods is quite
small. An important source of the relative scarcity of work on the service sector in general and
trade in services in particular is the fact that the service sector is not measured as well as the
goods producing sector.
The organizers of this conference asked me to 1) provide my perspective on whether measuring
the impact of trade in services is potentially important, 2) assess the prospects for measuring the
impact of trade in services, and 3) identify any data needs, provide priorities for the data needs,
and (somewhat provocatively) suggest organizational changes that might improve the statistical
system.
This is not new ground. Other organizations have produced reports on varying aspects of the
impact of outsourcing, offshoring, services trade, and data availability.1 I will not report on all
previous efforts, but will draw on the MIT/Sloan Offshoring Working Group report (Sturgeon
and Levy 2006), as I was a contributor and I think it still accurately reflects needs and priorities.
The purpose of this paper is to take stock of the current prospects for measuring the impact of
trade in services on the U.S. economy. I will describe progress economists have made over the
past 10–15 years using detailed, establishment-level microdata to examine the impact of trade in
goods on the manufacturing sector. I will argue that to investigate the impact (or potential
impact) of trade in services on the United States, one (or at least I) would want to use similar
methods.2 I will then describe what data would be needed to conduct this research and how much
of that data is currently available.
I then propose priorities for improving the ability of researchers to examine the impact of trade in
services on the U.S. economy. First, I provide a brief overview of developments in the U.S.
service sector.
1
Other organizations that have produced reports on this or related topics include the National Academy of Public
Administration, National Academy of Sciences, and Government Accountability Office.
2
This would be a good place to put my perspective in context. I am someone who has done microdata research
examining the impact of trade on the U.S.` manufacturing sector and tried to do the same for the service sector; not a
necessarily representative perspective but one that should support other types of analysis (aggregate data is only as
good as the microdata). So, while not everyone prefers to use microdata to examine these types of issues, conducting
similar studies on more aggregated data would require collecting the same information.
7
THE SERVICE SECTOR
Service Sector Employment
The service sector accounts for the lion’s share of employment in the United States (and most
other advanced economies). While services have traditionally been viewed as nontradable,
services trade is growing and there is an increasing sense that technological change is making it
easier and less expensive to provide services from a distance.
Table 1
NAICS Code
21
22
23
31-33
42
44-45
48-49
51-56
51
52
53
54
55
56
61-81
61
62
71
72
81
Sector
Mining
Utilities
Construction
Manufacturing
Wholesale trade
Retail trade
Transportation and warehousing
Business Services
Information
Finance and insurance
Real estate and rental and leasing
Professional, scientific, and technical services
Management of companies and enterprises
Administrative and support and waste remediation services
Personal Services
Educational services
Health care and social assistance
Arts, entertainment, and recreation
Accommodation and food services
Other services (except public administration)
Federal Government
State and Local Government
Employment
2007
703,129
632,432
7,399,047
13,333,390
6,295,109
15,610,710
4,435,760
33,430,809
3,428,262
6,562,546
2,249,353
8,121,171
2,915,644
10,153,833
34,595,857
562,210
16,859,513
2,070,524
11,587,814
3,515,796
2,462,000
16,400,000
Share of Total
Employment
2007
0.5%
0.5%
5.5%
9.9%
4.7%
11.5%
3.3%
24.7%
2.5%
4.9%
1.7%
6.0%
2.2%
7.5%
25.6%
0.4%
12.5%
1.5%
8.6%
2.6%
1.8%
12.1%
Employment
Growth
1997-2007
38%
-10%
31%
-21%
9%
12%
52%
29%
12%
12%
32%
51%
11%
38%
23%
75%
24%
30%
23%
8%
---
Table 1 presents information on employment and employment growth from the 1997 and 2007
economic censuses. Depending on the definition one uses, the service sector accounts for
between more than 60 percent to more than 80 percent of employment. Further, employment in
the service sector is growing, in contrast to the manufacturing sector. Services are a large and
growing part of the U.S. labor market.
Just because services are a large and growing portion of the U.S. economy does not necessarily
imply that trade in services is likely to affect the U.S. economy in a significant way. But the
confluence of a variety of changes (e.g., decreasing travel and telecommunication costs,
decreasing IT hardware costs, increasing Internet availability worldwide) seem to have
significantly increased the ease with which services are traded and expanded the scope of service
activities that can be provided at a distance. As a result, trade in services is growing.
8
Trade in Services: Official Statistics
Figure 1 shows the steady increase in U.S. service imports and exports. Both U.S. services
exports and imports about doubled between 1997 and 2007. Services now account for 30 percent
of U.S. exports and about 17 percent of U.S. imports.
Figure 1
US Services Trade 1997 - 2007
500
450
400
Billions of Dollars
350
300
250
200
150
100
50
0
1997
1998
1999
2000
2001
2002
Exports
2003
2004
2005
2006
2007
Imports
Source: Bureau of Economic Analysis.
Figures 2 and 3 show the composition of U.S. service exports and imports over the period 1992–
2007. While all of the categories exhibit growth, the “Other private services” category is growing
the fastest (both imports and exports more than doubling over the period) and contributes the
most to overall services import and export growth—Other private services account for more than
half of the overall increase in services exports and accounts for half of the increase in services
imports.
Other private services are comprised of the following activities: Education, Financial services,
Insurance services, Telecommunications, and Business, professional, and technical services.
Import and export data for these components of Other private services are only available starting
in 1997. For both imports and exports, the Business, professional, and technical services
category is the largest at the end of the period and contributes the most to Other private services
growth over the period. A long time series of the BPTS category is not available, so it is not
possible to decompose its growth into more detailed components.
9
Figure 2
Composition of US Service Exports
600,000
500,000
Millions of dollars
400,000
300,000
200,000
100,000
0
1992
1993
1994
Travel
1995
1996
Passenger Fares
1997
1998
1999
Other Transportation
2000
2001
2002
2003
Royalties and License Fees
2004
2005
2006
2007
Other Private Services
Figure 3
Composition of US Service Imports
400,000
350,000
Millions of dollars
300,000
250,000
200,000
150,000
100,000
50,000
0
1992
1993
1994
Travel
1995
1996
Passenger Fares
1997
1998
1999
Other Transportation
10
2000
2001
2002
2003
Royalties and License Fees
2004
2005
2006
Other Private Services
2007
Business, professional, and technical services; Financial services; and Insurance services account
for a significant share of service sector growth over the past 15 years. We would like to
understand better how trade in these types of services is affecting the United States. As discussed
in this section, the availability of detailed data going back in time poses a significant impediment
to researchers. And while the level of detail for trade in services data is improving, the level of
detail is still far more aggregated than in the manufacturing sector. The lack of historical data
and the ongoing lack of detailed industry-level data are two examples of the challenges in
measuring and analyzing trade in services. In the next section, I describe an alternative
methodology for assessing the potential scope of trade in services.
Tradable Services: A Different Perspective
Another less conventional (but more detailed) perspective on the potential for service trade to
affect the U.S. labor market is from work that Lori Kletzer and I (Jensen and Kletzer 2006) did
examining the tradability of service activities. We use the geographic concentration of service
employment across metropolitan areas within the United States to identify service activities that
are tradable. The intuition is that if services production is geographically concentrated (more
than demand for the service), it is probably being traded. As an example of this intuition, think of
personal services like haircuts or divorce lawyers. These service activities tend to be distributed
in proportion to the population in a region (and thus we don't see big concentrations of these
types of service activities in one place). But increasingly, there are services that don't seem to
require face-to-face interaction and thus might be tradable—think software development or
securities and commodities trading. We used this feature to distinguish between service activities
that are tradable and those that require face-to-face interaction (and thus are far less likely to be
traded).
We find that many service activities—such as movie and music recording production, securities
and commodities trading, software, and engineering services—appear to be traded within the
United States and thus are at least potentially tradable internationally. Approximately 14 percent
of the workforce is in service industries classified as tradable. By comparison, about 12 percent
of the workforce is in manufacturing industries classified as tradable. When workers in tradable
occupations (such as computer programmers in the banking industry or medical transcriptionists
in the health care industry) in nontradable industries are included, the share of the workforce in
tradable service activities is even higher.
While many service activities appear tradable, in related work (Jensen and Kletzer 2008), we
argue that only about one-third of the jobs in these activities will face meaningful competition
from low-wage countries (or risk being offshored) in the medium term. Tradable service jobs,
such as those at engineering or research and development firms, are good jobs. Workers in
tradable service activities have higher than average earnings. Part of this premium is due to
workers in these activities having higher educational attainment than other workers, but even
controlling for educational and other personal characteristics differences, workers in tradable
service activities have 10 percent higher earnings. Within the set of professional service
industries, a worker in a tradable industry and a tradable occupation has almost 20 percent higher
earnings than a similar professional service worker in a nontradable industry and occupation.
11
High earnings in tradable service activities do not mean that these jobs will be “lost” to lowwage countries. High-wage, high-skill activities are consistent with U.S. comparative advantage.
In the manufacturing sector, it is low-wage, labor-intensive industries like apparel that are most
vulnerable to low-wage import competition. The United States continues to have strong export
performance in high-wage, skill-intensive manufacturing industries.
The United States currently exports high-wage, high-skill services like computer software and
satellite telecommunications services. Most commentators on the offshoring issue focus on the
jobs that will be lost to offshoring but neglect that the United States has comparative advantage
in many service activities. Increased exports of services (and “inshoring”) are likely to benefit
many U.S. firms and workers.
About two-thirds of tradable business service jobs are skilled enough to be consistent with U.S.
comparative advantage. U.S. service workers and firms are likely to be beneficiaries of increased
trade in services through increased export opportunities.
The relationship between skills, wages, and trade highlights the need to have detailed data
covering the service sector.
MEASURING THE IMPACT OF TRADE IN SERVICES
How would researchers analyze the impact of trade in services on the U.S. economy? The
literature, both classical trade theory and more recent empirical and theoretical work, give us a
good idea of where to start. Traditional trade theory and more recent theoretical and empirical
advances suggest several important considerations: factor intensities and factor abundance,
productivity differences across countries, industries and within industries, and producer
heterogeneity within and across industries.
Lessons from Trade Theory
International trade theory emphasizes a number of features that help explain the sources and
implications of international trade. Traditional trade theory emphasizes that countries will trade
goods in which they have a comparative advantage—either through relative productivity
differences or through differences in relative factor endowments. Countries will tend to export
goods that they are relatively efficient at producing, either because they have a technological
advantage or because they are relatively abundant in the factor important to a good’s production.
Thus, capital-intensive countries like the United States tend to export capital-intensive products
and import labor-intensive products from countries where labor is abundant and wages are
relatively low.
These traditional trade theories also described the mechanism through which trade can affect
relative factor returns (i.e., wages and the returns to capital)—when countries specialize across
industries that differ in their use of different inputs, the relative returns to the inputs may change.
When industries that make relatively intensive use of unskilled labor (e.g., apparel production)
shrink, employment prospects and wages for unskilled workers are likely to suffer.
12
This traditional trade theory emphasized how differences across countries will influence the
patterns of trade. Yet, a large share of international trade takes place between relatively similar
trading partners, apparently within industries (see Grubel and Lloyd [1975]). Germany and the
United States, for example, exchange automobiles. This fact and others led to the creation of
“new” trade models that emphasize economies of scale in production and consumer preferences
for different varieties. In these models, otherwise similar firms (operating in countries with
similar factor endowments) specialize in different varieties, spurring two-way or “intra-industry”
trade between countries (see Ethier [1982], Helpman [1981], and Krugman [1980]).
Recent Lessons from Empirical Research in Manufacturing
One feature of both old and new trade theory is that the theories typically assume a
representative firm—that is, they assume all firms within an industry are the same. There is a
growing body of empirical research using plant- and firm-level microdata for the manufacturing
sector that demonstrates that the traditional assumption of a “representative” firm in an industry
is not appropriate for many research questions, including understanding the impact of
globalization.3 Plants, even within narrowly defined industries, exhibit considerable
heterogeneity both in their cross-sectional characteristics and in their behavior over time. The
heterogeneity of plants and firms and the variation in their responses to globalization have clear
implications for the impacts of trade in services.
Within Industry heterogeneity in manufacturing
Bernard and Jensen (1995) provide some of the first plant-level results on U.S. exporters and
find that exporters are relatively rare. Even in industries in which the United States has a
comparative advantage, the majority of plants do not export, while even in import-competing
sectors like textiles and apparel some firms export. In addition to being relatively rare, exporters
are strikingly different from plants in the same industry. Exporters are significantly larger than
nonexporters in the same industry. Exporters are also more capital intensive, more skilled-worker
intensive, and pay higher wages than plants of similar size, in the same industry, in the same
state. Exporters are also more productive than nonexporters in the same industry and region.
Bernard and Jensen (1999, 2006) also show that exporters are more likely to survive and have
higher employment growth than nonexporters of similar size, in the same industry, in the same
region. Because exporters have different characteristics than nonexporters, and because they
have differential growth and survival rates, the potential exists for the behavior of exporters to be
associated with 1) a reallocation of economic activity that affects aggregate measures like
industry and aggregate productivity and 2) the demand for and returns to different factors of
production (e.g., skilled workers).4
3
This section is not meant to be exhaustive or representative. For more comprehensive reviews see Bernard et al.
(2007), Helpman (2006), and Roberts and Tybout (1996), which focus on developing market contexts. Here, I draw
mostly on work I have been involved in to demonstrate how one might go about this type of research.
4
These relationships are not restricted to export participation. Researchers have also examined the characteristics
and behavior of multinational corporations using plant- and firm-level microdata. Doms and Jensen (1998) find that
U.S. manufacturing plants owned by MNCs—either foreign MNCs or U.S.-based MNCs—have superior operating
characteristics relative to domestic-owned plants. Bernard and Jensen (2007) explore the behavior of MNCs over
time and show that plants owned by U.S. MNCs are unconditionally more likely to survive, though controlling for
13
Economists are now incorporating these empirical regularities into models of international trade
and investment (for example, see Bernard et al. [2003]; Bernard, Redding, and Schott [2006];
and Melitz [2003]). While differing in their details, these models have several shared
implications. As trade costs fall, low productivity nonexporters are more likely to fail, highproductivity nonexporters are more likely to start exporting, and existing exporters should
increase their exports.
These models have direct implications for how increased trade will affect firms and workers. If
trade costs are reduced differentially across industries (either because of policy or technology),
industries with larger reductions in trade costs are likely to see more churning within the
industry. Because low-productivity plants tend to use low-skill and low-wage workers more
intensively, the increased likelihood of plant failure has implications for the demand for low-skill
workers. To the extent that particular industries or low-productivity producers are concentrated
in particular geographic areas, this will also affect distributional outcomes. In this section we
review results that examine the impact of international trade on U.S. manufacturers explicitly.
Competition from low-wage countries
Bernard, Jensen, and Schott (2006a) examine the role of import-competition from low-wage
countries on the reallocation of U.S. manufacturing within and across industries from 1977 to
1997. They focus on where imports originate (rather than their overall level), motivated by the
factor proportions framework and the significant increases in import shares from low-wage
countries like China. Their use of plant-level data provides a richer examination of U.S. producer
responses to international trade, including plant exit and product switching, than is possible with
more aggregate data. Specifically, their analysis identifies whether reallocation within industries
is consistent with U.S. comparative advantage.
They show that low-wage country import shares and overall penetration vary substantially across
both industries and time. Both components tend to be higher and to increase more rapidly among
labor-intensive industries such as Apparel and leather. Other industries such as Textiles see only
modest rises in both series. Finally, more capital- and skill-intensive sectors such as
Transportation and industrial machinery experience rapid growth of import penetration but little
or no increase in the share of imports from low-wage countries. They find that plant survival and
employment growth are negatively associated with industry exposure to low-wage country
imports. Within industries, they show that manufacturing activity is disproportionately
reallocated toward capital-intensive plants. Because there is an observed empirical regularity that
capital-intensive plants also tend to be more skill (nonproduction worker) intensive, the
reallocation to more capital-intensive plants will likely have implications for the relative demand
for skilled and unskilled workers.5
the superior operating characteristics of MNCs, MNC-owned plants are actually more likely to close. Firm
participation in international markets is significantly correlated with both plant characteristics and behavior over
time.
5
Bernard and Jensen (1997) examine the impact of reallocation to exporters on the relative demand for and wages
paid to skilled workers in the U.S. manufacturing sector.
14
Falling trade costs
In separate but related work, Bernard, Jensen, and Schott (2006b) examine the impact of falling
trade costs (both tariffs and transportation costs) on U.S. manufacturers. They find that when
trade costs in an industry fall, plants are more likely to close. This is one channel by which
international trade can affect the distribution of economic activity, aggregate productivity
growth, and the demand for labor. Falling trade costs tend to reduce the amount of economic
activity at the low end of the productivity distribution. This tends to raise aggregate productivity
(even without any technological change at the plant level) by truncating the low end of the
productivity distribution. Because low productivity plants also tend to be production-worker
intensive, this change is likely to affect the relative demand for unskilled workers.
They find that relatively high-productivity nonexporters in industries with falling trade costs are
more likely to start exporting. They also find that existing exporters increase their shipments
abroad as trade costs fall. Exporters are relatively high-productivity plants, and the expansion of
the high end of the productivity distribution will tend to raise aggregate productivity (even if no
plant changes its productivity). Because exporters are skill and capital intensive, this will also
tend to increase relative demand for these factor inputs. Bernard, Jensen, and Schott also find
that decreases in trade costs, and the increased competitive pressure associated with it, are
associated with increased productivity at the plant level. Not surprisingly, given the number of
channels by which falling trade costs shift the distribution of economic activity toward more
productive plants, they find that industries experiencing relatively large declines in trade costs
exhibit relatively strong productivity growth.
U.S. multinationals and outsourcing
Hanson, Mataloni, and Slaughter (2005) examine multinational behavior with regard to the
choice of the location of production using confidential data from surveys conducted of all U.S.
multinationals. They use a direct measure of input flows associated with vertical production
networks: foreign affiliates’ imports from U.S. parent firms (and other U.S. entities) of
intermediate inputs for further processing. They estimate the sensitivity of demand for imported
intermediates for additional processing to host-country and industry trade costs, factor prices,
taxes, and other variables suggested by theory.
Manson, Mataloni, and Slaughter find that imports of intermediate inputs are strongly negatively
correlated with trade costs facing affiliates. They find that vertical production networks are
sensitive to labor costs—imported-input demand is decreasing in host-country wages for lessskilled workers and increasing in host-country wages for more-skilled workers. They find that
foreign affiliates do more processing of imports in countries with relatively cheap, less-skilled
labor. A third finding is that vertical production networks also depend on other host-country
policies and characteristics. Imported-input demand is higher in host countries with exportprocessing zones, and is decreasing in host-country market size and corporate tax rates.
The examples of research described in this section demonstrate the usefulness of detailed,
comprehensive microdata in analyzing the impact of globalization. In the next section I describe
data needs to produce similar analyses for the service sector.
15
DATA NEED TO ANALYZE GLOBALIZATION IN THE SERVICE SECTOR
We know a considerable amount about the reaction of firms to changes in the global trading
environment in the manufacturing sector. If a researcher were interested in conducting similar
research on the service sector, what are the prospects?
To understand how increased trade in services might affect the U.S. economy, both theory and
previous empirical work stress some key considerations for understanding the impact:
1) The prevalence (how many activities?), scale (how much is being traded?), and direction
(who is trading with whom?)of trade in services.
2) How trade in services has evolved over time.
3) The factor intensities used in services provision.
4) The factor intensities across locations.
5) Firm-level heterogeneity (in size, factor intensities, productivity, trade activity) within
and across industries and countries.6
Measuring Trade in Services
The Bureau of Economic Analysis (BEA) collects information on trade in services and presents
aggregate data on international services transactions through three publication programs: 1)
cross-border trade in services data in the international transactions accounts; 2) sales of services
through affiliates of multinationals, some portion of which represent cross-border trade; and 3)
benchmark input-output tables.
The cross-border trade in services publication program provides the basis for all of the BEA’s
services trade data. As a result, this publication program provides the best sense of the trade data
the BEA collects:
The estimates of cross-border transactions cover both affiliated and unaffiliated transactions
between U.S. residents and foreign residents. Affiliated transactions consist of intra-firm
trade within multinational companies—specifically, the trade between U.S. parent companies
and their foreign affiliates and between U.S. affiliates and their foreign parent groups.
Unaffiliated transactions are with foreigners that neither own, nor are owned by, the U.S.
party to the transaction.
Cross-border trade in private services is classified into the same five, broad categories that
are used in the U.S. international transactions accounts—travel, passenger fares,
“other transportation,” royalties and license fees, and “other private services.”
(Survey of Current Business, November 2001)
6
While not exactly a data need, if researchers are to use information on producer heterogeneity, they need access to
producer-level information, i.e., microdata, which is often collected under a pledge of confidentiality. Thus, access
to producer-level microdata is an additional dimension of data needs.
16
Affiliated transactions are collected through the BEA’s U.S. Direct Investment Abroad and
Foreign Direct Investment in the U.S. programs. Comprehensive benchmark surveys are
collected every five years, and less comprehensive collections are conducted annually.
The BEA collects data on U.S. international transactions in private services with unaffiliated
foreigners through 11 surveys. These surveys fall into three broad categories: 1) the surveys of
“selected” services, which cover mainly business, professional, and technical services; 2) the
specialized surveys of services, which cover construction, engineering, architectural, and mining
services, insurance services, financial services, and royalties and license fees; and 3) the surveys
of transportation services. These collection programs are the principal source of the BEA’s
estimates of trade in services, but the estimates of some services are based on data from a variety
of other sources, including U.S. Customs and Border Protection and surveys conducted by other
Federal Government agencies, private sources, and partner countries.
Need: Increased detail—industry and country
Detailed data on international services transactions for cross border trade are currently available
from 1986 through 2006. Service imports and exports are reported for approximately 30 (1986–
1991) to 35 (1992–2006) service types (with additional detail on whether the transactions are
between affiliated or unaffiliated parties available for some categories). These data are available
by country for approximately 35 countries and country groupings for 1986–2006.
Figure 4 exhibits the detail on trade in services (both affiliated and unaffiliated) published by the
BEA over time. It shows the significant increase in detail over the past decade. The figure also
shows how large the gap is between the detail available for the manufacturing sector (where
information is available for over 8,000 export categories and over 10,000 import categories) and
the service sector. The published aggregates are moving in the right direction, but we clearly
have a ways to go.
17
Figure 4 – Categories reported in the BEA Table 1b 1992–2006
1992
1997
Travel 2 …………………………………… Travel 2 ……………………………………………………………………...
2001
2006
Travel 2 ……………………………………………………………………...
Travel 2 ……………………………………………………………………...
Passenger fares 3 ……………………… Passenger fares 3 ………………………………………………………….
Passenger fares 3 ………………………………………………………….
Passenger fares 3 ………………………………………………………….
Other transportation…………………… Other transportation……………………………………………………..
Other transportation……………………………………………………..
Other transportation……………………………………………………..
Royalties and license fees……………… Royalties and license fees………………………………………………
Royalties and license fees………………………………………………
Royalties and license fees………………………………………………
Other private services 4
15
……………… Other private services 4
Education 5……………………………
15
…………………………………………………..
15
…………………………………………………..
15
…………………………………………………..
Education 5………………………………………………………………..
Education 5………………………………………………………………..
Education 5………………………………………………………………..
Financial services 16……………………………………………………….
Insurance services 6 …………………
Insurance services 6 ……………………………………………………
Telecommunications 7…………………
Telecommunications 7………………………………………………….
Business, professional, and technical services 16…………………
Computer and information services 8 16……………………………….
9
Operational leasing 16………………………………………………………
Other business, professional, and technical services
10 16
…………..
Management and consulting services ………………………………
Management and consulting services 9………………………………
Research and development and testing services 9………………..
Research and development and testing services 9………………..
Other business, professional, and technical services
10 16
…………..
Other business, professional, and technical services 10 16…………..
Accounting, auditing, and bookkeeping services………………..
Advertising……………………………………………………………
Architectural, engineering, and other technical services………..
Construction ………………………………………………………….
Industrial engineering……………………………………………………
Installation, maintenance, and repair of equipment…………………
Legal services……………………………………………………………
Medical services 11…………………
Medical services 11…………………………………………………………
Mining 12……………………………………………………………………
Sports and performing arts……………………………………………..
Trade-related services 13………………………………………………..
Training services…………………………………………………………
Other 14…………...
Other services…………………………
Other services…………………………………………………………...
Other services…………………………………………………………...
Other services…………………………………………………………...
Film and television tape rentals…
Film and television tape rentals……………………………………….
Other…………………………………
Other…………………………………………………………………………
Other…………………………………………………………………………
Other…………………………………………………………………………
18
“What is most troubling for us is that the seventeen industry categories listed in the first
column of Table 4 exhaust the detail on services trade collected by United States
government statistical agencies. What is going on in the other service product categories
that have been mentioned as moving offshore, such as the wide variety of back-office
functions like accounting, customer support, and software programming? What about the
interpretation of radiology images, market and legal research, and research to support
financial services? Are customized software services staying onshore while only basic
software coding is moving offshore, or is higher-skilled work and work related to
innovation and new product creation also being imported? Because very few questions
are asked, very little detail is collected, leaving us with extremely thin data on services
trade, even if steps are taken to improve data quality. Contrast the seventeen descriptive
categories for traded services products in Table 4 with the more than 16,000 detailed
product codes for goods collected by the United States Department of Commerce and the
magnitude of the data gap becomes clear. It is clearly infeasible to collect as much
product detail on services trade as is generated by the customs forms filled out when
goods are shipped across borders. But much more detail could and should be collected.”
(Sturgeon and Levy 2006)
Progress is being made. The BEA has resolved the inconsistency between the survey formats for
affiliated and unaffiliated trade. This now permits greater detail in reporting the types of services
traded. While this represents progress, it does not resolve the issue of the need for greater detail.
Need: Lower reporting thresholds
“While the BEA surveys that ask firms to quantify their trade in services are mandatory,
firms are exempted from reporting categories of services in which they have import
transactions of less than $6M per year and export transactions of less than $8M per year.
In the case of services, in particular, because firms tend to be smaller than firms engaged
in goods trade, the current thresholds very likely exclude many transactions. Because of
this, we believe that the thresholds for mandatory reporting of international services
transactions should be lowered.”
Need: Increased sample/improved sampling frame
Related to the issue of lowering reporting thresholds, the BEA needs to improve its capacity to
develop survey frames.
“Another explanation for the apparent undercounting of services trade is that the BEA is
not collecting data from the right companies, or is sending inappropriate surveys to the
companies on its mailing lists. To test for potential undercounting of U.S. services
imports, the Government Accountability Office (GAO) provided the BEA with a list of
104 firms identified from press and company reports as likely to be importing services
from India. The BEA was asked to compare this list with the survey responses it had
received from firms on its mailing lists. The BEA had 87 (84 percent) of the firms
19
identified by the GAO on its mailing lists. The BEA stated that it had dropped some of
the missing companies from its mailing lists because they had not previously met the
reporting thresholds for services trade.”
“Furthermore, only 54 (52 percent) of the firms identified by the GAO had received
appropriate surveys from the BEA (e.g., firms with offshore affiliates were not sent the
survey on affiliated trade). Finally, only 15 (14 percent) of the 104 firms identified by the
GAO as likely to be importing services from India reported such imports (GAO 2005b, p
19). One explanation for the low level of reporting of services trade with India is that
firms that had transactions valued beneath the thresholds mentioned above, while not
required to do so, nevertheless filled out the BEA surveys but did not provide detail on
the source or destination countries associated with their international transactions because
they were not required to do so.”
“Still, the BEA believes that its data on services trade is of good quality. When the BEA
contacted the companies on the GAO list that were missing from its mailing lists, it did
not identify any company with substantial imports of services that were not already being
reported. Nevertheless, the BEA recognizes that more resources need to be allocated
toward maintaining lists of survey respondents since the identity of transactors may
change from year to year. The BEA has a variety of initiatives underway to improve its
mailing lists and improve survey compliance (GAO 2005b, p. 20). The BEA also plans to
merge the collection of its data on affiliated international services transactions with its
data on unaffiliated international services transactions, so that a given type of service is
covered in exactly the same detail, whether it is imported or exported, and whether it is
with an affiliated or an unaffiliated foreign party. We believe that these efforts are
significant and very helpful, especially if combined with lower thresholds for mandatory
survey compliance.”
The BEA is now collecting information from on unaffiliated and affiliated international service
transactions using the same collection form. This resolves the issue of the information being
collected at different levels of detail.
The BEA has undertaken efforts to improve its sampling frame. It commissioned the Census
Bureau to add a question to the 2006 Company Organization Survey to ask whether firms
imported services. The purpose of this additional question is to improve the sampling frame for
the BEA’s data collection programs.
20
Measuring the Impact of Services Trade on the U.S. Economy
To understand how increased trade in services is likely to affect the U.S. economy, requires the
detailed information on trade flows described above and the ability to link it to detailed
information on domestic producers. Specifically, I would want detailed information on the inputs
service firms use (labor, capital, land, buildings, accounting services, intellectual property, etc.)
and the outputs they produce (computer programs, lawsuits, ad campaigns, medical operations,
etc.). These data would help me understand the relationship between growth in demand for
particular services and the demand for inputs to those services. These data would also help me
understand whether productivity within the service sector is increasing over time (and whether
this growth is in response to particular changes in the environment). To understand how the
service sector affects employment outcomes across regions within the United States, I would
want these data on a (hopefully detailed) geographic basis. I would also need to be able to link
these data to detailed information on international trade in services (the type of information
discussed above).7
Need: More detailed industry classification
The data covering the service sector within the United States are not as robust as the data for the
manufacturing sector in a number of dimensions. The information collected from the service
sector—for both inputs and outputs—is less detailed. A simplistic example of how output in the
service sector is not collected at as detailed a level as the manufacturing sector is looking at
NAICS codes per worker across sectors in the economy.8 NAICS contains about 470 industrial
codes for the manufacturing sector (NAICS 31–33). For the service sector (NAICS 51–81),
NAICS contains about 325 industry codes. The manufacturing sector employed about 13 million
people and the service sector employed about 68 million workers in 2007. In terms of workers
per industry code, there were about 28,000 workers per NAICS code in manufacturing in 2007
and about 208,000 workers per NAICS code in the service sector. By this crude metric, the
service sector is substantially underclassified (almost 10 times so).
While the number of industries in the service sector relative to the manufacturing sector is low,
the implementation of NAPCS is improving the level of detail for the output of establishments in
the service sector. The 2007 economic census forms for the service sector have considerable
detail for output product categories within service industries.9
7
As described above, this type of data is available for the manufacturing sector. The Census Bureau and made
available publicly in aggregated form and made available in disaggregated form to approved researchers at the
Center for Economic Studies. The research community has learned a great deal about the manufacturing sector
across a wide range of topics—productivity dynamics, job creation and destruction, impact of environmental
regulation, and impact of trade, just to name a few—through access to producer level information at the Census
Bureau.
8
While this is not necessarily the only (or best) way to think about classification, if one is interested in labor market
impacts it is instructive to note the significant difference in the industry detail available across sectors.
9
While this is helpful, an issue with classifying establishments into broad industries and collecting detailed product
information is that it is difficult to allocate inputs across outputs. Additional refinement of the service sector industry
codes would improve the ability to measure things like productivity.
21
Need: More detailed information on inputs to the production process
Another way in which the service sector data are less robust than the manufacturing sector is
with regard to the collection of data on inputs into the production process.
“The Census Bureau has developed detailed classification schemes for material inputs
and manufactured products that it uses to collect information on what individual
manufacturing establishments buy and sell. These product categories have been
developed with a great deal of care, and government surveys have been tuned to specific
sectors. For example, establishments in the plastics industry are required to provide
detailed information about the consumption of chemical feedstock and the production of
various kinds of plastics while establishments producing furniture are required to provide
detail about the consumption of wood, metal, hardware, glue, and fabric and the
production of various kinds of furniture. This pattern holds true across the manufacturing
sector. The U.S. Census Bureau’s Numerical List of Manufactured and Mineral Products
contains hierarchically organized descriptions of the principal products and services of
the manufacturing and mining industries in the United States. These codes are used to
collect data for the Economic Census and are used by the Bureau of Economic Analysis
for the input-output matrix that underlies the national accounts. But as in international
trade in services, far less detail is collected on the services products that are consumed
and produced domestically. Again, there are more than 6,000 codes for physical products
but fewer than 100 for services.
The lack of detail on domestic trade in services means that the Bureau of Economic
Analysis largely estimates the contribution of services to the national accounts. While
resulting estimation cannot claim precision, BEA analysts believe that their techniques
capture the magnitude and direction of change in services accurately enough to support
policy. While this may be true today, we think the view of the U.S. Census Bureau,
quoted in full in the previous section, bears repeating, “If [the information gap between
manufacturing and services goes] unaddressed, economic policymakers will be
increasingly misinformed and misdirected about changes in the real economy, related to
rates and sources of growth in output, prices, productivity, and trade.” Clearly, an
accelerated and sustained effort to collect more detail on domestic trade in services is
required. Our second recommendation, therefore, is for the U.S. Census Bureau to
accelerate the completion the North American Product Classification System (NAPCS),
and fully and rapidly deploy it in the Economic Census, at the establishment level, for
both inputs and outputs.” (Sturgeon and Levy 2006)
The recommendation above is with regard to purchased inputs used to produce services. I think
this is an important improvement that would be beneficial to helping to understand how the
service sector functions.
In addition to increased information on purchased services, I would like to suggest two other
improvements. We learned from the literature on the impact of trade on the manufacturing sector
that factor intensities (both capital intensity and skill intensity) are important determinants of
22
how establishments behave in response to international competition. It would be useful if the
Census Bureau would collect information on the skill intensity of the workers that are employed
in the service sector. Currently, the economic censuses do not consistently collect information on
labor inputs other than total employment and salaries and wages.10 It would be beneficial if the
Census Bureau collected more information than just total employment and wages. I recognize
that detailed information on skills or educational attainment would be costly to collect and
burdensome to provide. However, I think that the research in the manufacturing sector
demonstrates that it is possible to collect very crude classifications (in the case of manufacturing
production and nonproduction workers) that still provide important information regarding the
skill intensity of firms’ production processes.
For services, the production/nonproduction worker classification might not make sense, but an
analogous classification might be exempt and nonexempt employees.11 While not an ideal
measure of skill, this classification is likely to capture meaningful variation in skill intensity
across producers and industries. It would be relatively easy to collect and probably relatively
straightforward for firms to report.
Another input that has proved to be an important determinant in plant survival in the
manufacturing sector is capital intensity. Currently, the economic censuses do not consistently
collect capital information. While it might not be particularly meaningful for some service
industries, for others it is not difficult to imagine that capital intensity would have something to
do with firm performance. (One can imagine that capital intensity of hospitals would be
systematically related to outcomes and, perhaps, likelihood of participating in international
trade.)
Need: Information on a geographic basis
The Census Bureau does collect information on capital expenditures in the Annual Capital
Expenditure Survey (ACES); however, ACES is an enterprise-level survey. Because many large
firms by employment and output operate in multiple industries and multiple geographic markets,
enterprise-level information on capital expenditures makes allocating capital service inputs to
locations and industries difficult. This highlights another desirable feature of information on the
service sector—geographical information.
To understand how international trade is affecting regions within the United States, it is
important to be able to examine how producers in different regions may vary in factor intensity
and productivity. This need highlights the importance of collecting as much information as
possible at the establishment level.
Collecting information at the establishment level enables researchers to place the economic
phenomena in a region and also enables a much tighter alignment of inputs used and outputs
(industries/products). Collecting information at the enterprise level seriously reduces the level of
10
For some industries, the censuses collect information on the type of worker (by training or activity, give examples
from engineering, lawyers, doctors’ offices).
11
Employees whose jobs are governed by the Fair Labor Standards Act (FLSA) are either “exempt” or
“nonexempt.” Nonexempt employees are entitled to overtime pay.
23
product and geographical specificity of the data. For some purchased inputs (e.g., advertising) it
may be difficult to collect the information at the establishment level. Yet, for inputs like physical
capital, it seems feasible to collect capital stock and flow data at the establishment level. (Capital
stock information is collected in the census of manufactures.)
Need: Researcher ability to access ad link microdata
As I described above, research using microdata provides a better understanding of how
globalization affects the U.S. economy. Researchers need access to microdata to conduct this
type of research.
“Steps should be taken to extract as much information as possible from the data that is
currently collected by government programs. An inventory of current and potential
microdata resources should be made, and as many “micro-data” sets as possible should be
archived, maintained, and made available to both government and academic researchers.
Micro-data are the data that supports government administrative programs and underlies
published statistics. In general, quantitative research based on micro-data can provide a
better and more detailed view of services offshoring and its effects than research based on
published statistics.”
A minor note related to microdata access is the desirability of permitting researchers to combine
data that has already been collected in different agencies to answer important questions. This is a
cost effective way of increasing the usefulness of data that has already been purchased.
“Finally, it is important to encourage research that links various sets of micro-data. While
there can be legislative and institutional barriers to sharing micro-data across agencies,
reducing these barriers could enable some extremely powerful research. For example if
the outbound foreign affiliate investment collected by the Bureau of Economic Analysis
in its surveys of multinational firms were to be combined with the firm, establishment,
and trade data collected by the U.S. Census Bureau, it would help researchers create a
more comprehensive picture of the operations of U.S. firms—both at home and abroad.
The combined data could reveal domestic activity at the establishment level (with product
level information, geographic information, and export information), the relationship
between the establishments within the firm, the amount of trading the firm does (using
the matched transaction and firm data), and the nature of the firm's foreign affiliate
operations (employment, wage bill, location, local sales, trade with parent, etc). This
would allow researchers to examine the relationship between domestic activity, trade, and
foreign direct investment.”
I understand the need to protect the confidentiality (and the perception of confidentiality) of
respondent level information. My strong sense is that the protocols and infrastructure necessary
to protect the confidentiality and perception of confidentiality are in place to restrict access to
approved uses within the Census Bureau, the BEA, and the BLS. It is my sense, however, that
24
bureaucratic impediments continue to impede researchers’ ability to combine and link datasets
from different statistical agencies.
IMPEDIMENTS TO IMPROVEMENT
In this section I describe what I perceive as impediments to improving the quality of data needed
to evaluate the impact of trade in services on the U.S. economy.
Resource Issues
As described in the first section of the paper, services are a large, important, and growing sector
of the U.S. economy. Yet, the infrastructure for collecting information on the service sector is
not as robust as that for other sectors like manufacturing. A primary reason for this disparity is
that Congress does not allocate the same level of resources (proportional to the size of the service
sector) as it does to the manufacturing or other sectors. Given this, it should not come as a
surprise that one impediment to improving statistics on trade in services and domestic service
activity is the need for additional resources.
As a simple metric of the disparity in resources devoted to the various sectors, the table below
shows the FY 2009 budget for the economic census by sector. I also show the number of
employees and the number of establishments in each sector. I then calculate the budget dollars
per employee and per establishment across sectors. The table shows that the resources devoted to
the service sector on a per employee basis or per establishment basis are significantly lower than
those devoted to manufacturing or mining.
On a per establishment basis, Congress allocates more than six times more money for data
collection in the manufacturing sector than in the service sector. On a per employee basis, the
disparity is smaller, but still more than twice is much is spent per employee in manufacturing
than in the service sector. If one compares mining, the disparities are even greater.
Economic Census Program Components (dollars in millions) FY 2009:
U.S. Census Bureau Data Collection
Sector
FY 2009
Budget
2007
Employment
2007
Establishments
Budget per
Employee
Budget per
Establishment
(millions)
Services
$39.9
68,026,666
4,382,720
0.59
9.10
Retail Trade
$23.7
15,610,710
1,122,703
1.52
21.11
Manufactures
$17.8
13,333,390
293,919
1.33
60.56
Wholesale Trade
$12.6
6,295,109
432,094
2.00
29.16
$6.8
$3.1
$1.7
7,399,047
5,068,192
703,129
725,101
234,805
21,169
0.92
0.61
2.42
9.38
13.20
80.31
Construction
Transportation, Communication, and Utilities
Minerals
Note: These numbers represent the budget for FY 2009. Not all periodic census activity associated with the Economic Census
occurs in FY 2009. However, because the timing of the processing for the various sectors within the Economic Census is similar, I
am assuming that the relative size of the budgets is representative of the total costs associated with each sector.
Source: U.S. Census Bureau, Periodic Censuses and Programs Budget Amendment FY 2009, as presented to Congress June
2008, Exhibit 12
25
This is a simple (maybe simplistic) metric, but it makes the point that service sector data
collection is relatively resource poor. To bring the data available for the (domestic) service sector
to a similar level as the data available for the manufacturing sector will require a commensurate
investment of resources.
To provide information on trade in services comparable to the information on trade in goods
does not seem feasible because goods pass through ports and are required to file customs forms
or shippers export declarations. These administrative systems provide a relatively inexpensive
means for collecting very detailed information on trade in goods. Because traded services do not
necessarily pass through ports, there is no obvious low-cost data collection system. It seems
likely that collecting information on trade in services will require survey responses from firms.
This is obviously more expensive than piggy-backing off administrative systems. To collect
better information on services, trade will at a minimum require a significant investment of more
resources. In the next sections, I describe what I perceive as additional prerequisites for
collecting better trade in services data.
Sampling Frame
An issue identified in the MIT Offshoring Working Group report is that the BEA does not have
access to an adequate sampling frame for conducting its surveys of international service
transactions. The BEA recognizes the need to improve its sampling frame and is, as described
above, taking steps to do so. Yet, I think it remains an open question of whether these modest
steps to improve the sampling frame are sufficient. What the BEA needs is access to a sampling
frame similar to that maintained by the Census Bureau.
Data-sharing legislation provides authorization for the statistical agencies to share confidential
data, but the situation is complicated by the fact that the Census Bureau’s business sampling
frame contains federal tax information provided by the Internal Revenue Service. For the Census
Bureau to share its sampling frame with the BEA or the BLS, companion legislation would have
to be passed that would amend section 6103(j) of Title 26 (governing the use of federal tax
information). This companion “j-bill” has not passed. If the Census Bureau could provide
sampling frame information to the BEA, it would be a significant improvement in the BEA’s
capacity to conduct surveys. I do not know what the current thinking is on the prospects for
passage of the companion “j-bill,” but evidence to date leaves one less than optimistic about
passage.
As a result of the lack of an adequate sampling frame, resource constraints, and the fact that the
principal mission of the BEA is to produce aggregate economic accounts, the BEA focuses its
data collection efforts on large organizations that they deem to be likely to trade services. My
impression is that the international transaction surveys are not statistically representative samples
across service sector industries, firm size classes, or geography. To improve the level of detail
available for trade in services statistics, the BEA will need to increase the number of
organizations it surveys and, presumably, increase the statistical representativeness of the
sample. These will require access to an adequate sampling frame.
26
Organization Structure
The conference organizers asked that I give some thought to organizational changes that might
facilitate improvements in service sector data. This is a potentially provocative topic, so I
approach it with some trepidation. Yet, if one takes a step back and looks at the organizational
structure for the collection of trade in services data, the choice of organization across agencies is
striking. The BEA is a recipient of large amounts of data collected by other statistical agencies
(including the BLS and the Census Bureau). It is also a data collection agency. In contemplating
this, I was left with the question: Why does the BEA collect information on multinational
enterprises and international service transactions?
While not based on much historical research, it is my impression that trade in services statistics
have historically been collected largely to fulfill the needs of national income and product
account (NIPA) construction. Other types of production and international trade data are collected
for a broad range of uses (including and importantly for the NIPAs). Historically, there has not
been large demand for detailed trade in service statistics beyond the need to complete the NIPAs.
I imagine that as a result of this feature of the data need, it made sense for the BEA to collect the
trade in services data.
Yet, I think this is beginning to change. As services share of the U.S. economy increases and
trade in services grows, there will be an increasing need to analyze the impact of a broader range
of phenomena associated with increased trade in services (e.g., the regional implications within
the United States, the impact of the service components of trade agreements).
As I argue in this paper, the need to understand the impact of trade in services—from a variety of
perspectives, e.g., impact of trade agreements, exchange rate impact, impact on local and
regional economies—require much more detailed data regarding trade in services. Researchers
and policymakers need comprehensive data across detailed industry classifications and
geographical regions within the United States—ideally not only which firms participate in global
services trade, but also which firms don’t. The data should be consistent with other productionrelated data and easily linked to other production data.
Collecting detailed, statistically representative information on trade in services across detailed
industries, countries, and regions within the United States is a major undertaking. An open
question is whether the BEA is the most appropriate agency to conduct the data collection.
There may be reasons why it makes sense to have a dedicated statistical agency within the BEA
for collecting this type of information. However, I see some significant drawbacks for this type
of fragmented collection system.
The first drawback is that data collection has fairly significant fixed costs—especially with
regard to developing and maintaining a sampling frame. As described above, the BEA’s inability
to access an adequate sampling frame is a significant impediment to improved trade in services
data collection. While I would not present myself as an expert in data collection methods, I can
27
imagine other examples of fixed costs in data collection (e.g., forms design expertise, survey
processing and follow-up capacity).
So, I think the big institutional question is why does the BEA collect these data? As identified
above, the lack of a proper sampling frame poses a significant impediment to the BEA’s ability
to carry out a statistically representative sampling of trade in service activity.
Another drawback is data consistency and potential problems with data integration. As an
example, when the BEA and the Census Bureau were directed to produce statistics at the
establishment level on foreign direct investment in the United States, the data comparability and
matching issues were not insignificant. If the foreign direct investment surveys and international
service transactions surveys were collected by the Census Bureau using the bureau’s sampling
frame and industrial and geographic coding systems, it would significantly increase the ease with
which the data could be used in conjunction with other production data.
There may be advantages to having the BEA conduct the survey that I am not aware of. The
BEA and the Census Bureau work closely on other aspects of data collection for the NIPAs. The
bureau has the infrastructure to collect detailed, statistically representative statistics on trade in
services. For example, it has arguably the best sampling frame for this type of application within
the statistical system. The Census Bureau already surveys all the relevant firms and
establishments. It appears to me that the efficiencies in data collection and improvements in
comparability from having these data collection activities within the bureau are potentially
significant. The costs and benefits of moving the foreign direct investment and international
service transactions data collection programs to the Census Bureau should be investigated.
28
APPENDIX A
SLOAN OFFSHORING WORKING GROUP RECOMMENDATIONS
Our working group had two purposes: 1) to evaluate the data available for characterizing and
measuring services offshoring and its effects on the United States economy, and 2) to make
recommendations for improvements in data collection, dissemination, and analysis.
We see three broad solutions to this problem, each of which should be aggressively pursued: 1)
more and better data on services trade should be collected, 2) more information should be
extracted and published from existing data resources, and 3) quantitative research methods
should be combined with qualitative methods to provide a better view of the context and
character of services offshoring.
Our five recommendations are as follows:
1) Collect more detail on international trade in services.
The BEA should collect more detail on services products that are traded internationally
(affiliated and unaffiliated services imports and exports). It currently collects data on only
17 categories of traded services products. In contrast, import and export statistics for the
United States are currently available for more than 16,000 categories of goods. Without a
more detailed view of which services are traded internationally, it will remain impossible
to determine which sectors experience pressure from import competition. As a result, we
cannot know where in the economy to look for the effects of services offshoring with any
precision. This in turn renders other data on services less useful.
2) Collect more detail on domestic trade in services.
The U.S. Census Bureau should accelerate its efforts to collect more detailed statistics on
services traded within the United States (services inputs and outputs). These more
detailed statistics will help to provide a better view of the role that services play in the
economy of the United States. Services account for more than 85 percent of U.S. private
sector GDP, but we have very little information on the services that are bought and sold
by companies.
3) Collect more detail and publish time series data on employment by occupation.
Because service work plays a role in all industries, adequate data on employment by
occupation is necessary to determine the employment and wage effects of services
offshoring. Data should be collected at the establishment level to enable links to data on
domestic and international trade. We recommend two concrete steps in this regard:
1) The BLS should publish consistent time series on employment by occupation
from the Occupational Employment Statistics (OES) program. If possible these
data should be published by industry at the national, state, and metropolitan
29
levels. Time-series data will allow policymakers to track employment trends in
the occupations most vulnerable to job loss from services offshoring.
2) The BEA should collect data on more occupational categories in its surveys on
the activities of U.S.-based multinational firms. More detail on the occupations
created by multinational firms, at home and abroad, will provide a clearer picture
of the employment effects of services offshoring.
4) Archive and provide access to more microdata resources.
Steps should be taken to extract as much information as possible from the data that is
currently collected by government programs. An inventory of current and potential
microdata resources should be made, and as many microdata sets as possible should be
archived, maintained, and made available to both government and academic researchers.
Microdata are the data that supports government administrative programs and underlies
published statistics. In general, quantitative research based on microdata can provide a
better and more detailed view of services offshoring and its effects than research based on
published statistics.
5) Accelerate research that combines quantitative and qualitative research methods.
No single approach or dataset can hope to bring the complex and dynamic phenomena of
services offshoring into complete focus. An interdisciplinary, collaborative approach is
needed to combine insights from data collected by government programs with insights
from researcher-generated surveys and field interviews. Quantitative methods allow
researchers to estimate the magnitude and speed of economic change and to implement
causality tests, while qualitative methods can provide a rich and nuanced picture of the
complexity, context, and dynamics of services offshoring.
30
References
Bernard, Andrew B., Jonathan Eaton, J. Bradford Jensen, and Samuel S. Kortum. 2003. “Plants
and Productivity in International Trade.” American Economic Review 93(4): 1268–1290.
Bernard, Andrew B., and J. Bradford Jensen. 1995. “Exporters, Jobs and Wages in U.S.
Manufacturing, 1976–87.” Brookings Papers on Economic Activity: Microeconomics.
———. 1997. “Exporters, Skill Upgrading, and the Wage Gap.” Journal of International
Economics 42: 3–31.
———. 1999. “Exceptional Exporter Performance: Cause, Effect, or Both?” Journal of
International Economics 47(1): 1–26.
———.. 2007. “Firm Structure, Multinationals, and Manufacturing Plant Deaths.” Review of
Economics and Statistics 89(2): 193–204.
Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott. 2007. “Firms
in International Trade.” Journal of Economic Perspectives 21(3): 105–130.
Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. 2006a. “Survival of the Best Fit:
Competition from Low Wage Countries and the (Uneven) Growth of U.S. Manufacturing
Plants.” Journal of International Economics 68(1): 219–237.
Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. 2006b. “Trade Costs, Firms, and
Productivity.” Journal of Monetary Economics 53(5): 917–937.
Bernard, Andrew B., Stephen Redding, and Peter K. Schott. 2007. “Comparative Advantage and
Heterogeneous Firms.” Review of Economic Studies 74: 31–66.
Doms, Mark E., and J. Bradford Jensen. 1998. “Comparing Wages, Skills, and Productivity
Between Domestic and Foreign Owned Manufacturing Establishments in the United
States.” In Geography and Ownership as Bases for Economic Accounting, Robert E.
Baldwin, Robert E. Lipsey, and J. David Richardson, eds. Cambridge, MA, and Chicago:
National Bureau of Economic Research and University of Chicago Press, pp. 235–255.
Ethier, William J. 1982. “National and International Returns to Scale in the Modem Theory of
International Trade.” American Economic Review 72(3): 389–405.
Grubel, Herbert G., and Peter J. Lloyd. 1975. “Intra-Industry Trade: The Theory and
Measurement of International Trade in Differentiated Products.” Economic Journal
35(339): 646–48.
Hanson, Gordon H., Raymond J. Mataloni, and Matthew J. Slaughter. 2005. “Vertical Production
Networks in Multinational Firms.” Review of Economics and Statistics 87(4): 664–678.
Helpman, Elhanan. 1981. “International Trade in the Presence of Product Differentiation,
31
Economies of Scale, and Monopolistic Competition: A Chamberlin–Heckscher–Ohlin
model.” Journal of International Economics 11(3): 305–340.
———. 2006. “Trade, FDI and the Organization of Firms.” Journal of Economic Literature
44(3): 589–630.
Jensen, J. Bradford, and Lori Kletzer. 2006. “Tradable Services: Understanding the Scope and
Impact of Services Offshoring.” In Offshoring White-Collar Work—Issues and
Implications, Brookings Trade Forum 2005, Lael Brainard and Susan M. Collins, eds, pp.
75–134.
———. 2008. “‘Fear’ and Offshoring: The Scope and Potential Impact of Imports and Exports
of Services.” Peterson Institute for International Economics Policy Brief 08-01.
Washington, DC: Peterson Institute for International Economics.
Krugman, Paul. 1980. “Scale Economies, Product Differentiation, and the Pattern of Trade.”
American Economic Review 70(5): 950–959.
Melitz, Marc J. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity.” Econometrica 71(1695–1725).
Roberts, Mark J., and James R. Tybout. 1996. What Makes Exports Boom. Washington, DC:
World Bank.
Sturgeon, Timothy J., and Frank Levy. 2006. “Why We Can’t Measure the Impact of Services
Offshoring: The Data Gaps and How to Fill Them.” Services Offshoring Working Group Final
Report. Cambridge, MA: MIT, Industrial Performance Center.
http://web.mit.edu/ipc/publications/pdf/IPC_Offshoring_Report.pdf (accessed May 19, 2010).
32
Measuring Success in the Global Economy: International
Trade, Industrial Upgrading, and Business Function
Outsourcing in Global Value Chains
An essay in memory of Sanjaya Lall
Timothy J. Sturgeon and Gary Gereffi*
September 30, 2009
This paper is published by UNCTAD in the journal Transnational Corporations 18:2,
(UNCTAD/DIAE/IA/2009/10), pp: 1-35 and can be found in its original format
on: http://www.unctad.org/templates/webflyer.asp?docid=12891&intItemID=2926&lang=1
This article contributes to an assessment of the scholarly work of Sanjaya Lall, especially as it
relates to improved measures of industrial upgrading and technological learning. We argue for
the collection of new statistics, in addition to reworking and linking existing datasets. Changes in
the global economy, especially the rise of global value chains (GVCs), have created
measurement problems that require not only continued innovation in the use of existing data
sources, but also the development and deployment of new measures that analyze GVCs more
directly. Specifically, we advocate for the collection of establishment-level economic data
according to business functions. Data collected according to a standardized set of generic
business functions can provide researchers and policy-makers with a better map of the value
*
Timothy J. Sturgeon is Senior Research Affiliate at the MIT Industrial Performance Center. Contact:
[email protected] . Gary Gereffi is Director of the Center on Globalization, Governance & Competitiveness at
Duke University. Contact: [email protected]. We thank Industry Canada for its support of an early draft of this
paper. We would also like to thank Rajah Rasiah for organizing these special issues and for his editorial work. Two
anonymous reviewers provided helpful suggestions. We owe Peter Bøegh Nielsen of Statistics Denmark our
gratitude for providing us with data from the European Union Survey on International Sourcing, and for his patience
and generosity in helping us with its presentation and interpretation. We acknowledge Ursula Huws and Sharon
Brown for their path breaking efforts and tireless enthusiasm for the business function approach. Most importantly,
we remember the late Sanjaya Lall for his inspiration, warm collegiality and sparkling intelligence.
33
chain, reveal the roles that domestic establishments, firms, and industries play within GVCs, and
offer a unique view of the competitive pressures facing domestic firms and industries.
Keywords: global value chains, international trade, business function outsourcing, industrial
upgrading, technological learning
Published in Transnational Corporations, Vol. 19, No. 2, August, 2009
34
This article contributes to an assessment and celebration of the scholarly and policy work of the
late Sanjaya Lall. As Rasiah (2009) highlights, Lall’s work was at once broad, deep, and
intensely focused. Over his long career, Lall and his many collaborators used the lenses of the
transnational corporation (TNC), competitiveness, globalization, and technological learning to
uncover the determinants of economic change—or lack thereof—in the developing world. There
is a clear continuity to this intellectual path, one that reveals Lall’s commitment to empirical
investigation, his skepticism of conventional wisdom, his openmindedness, and his sustained
focus on improving the lot of those in the world who have less.
During his early career, a time when TNCs were driving rapid economic development in pockets
of the developing world, he did not simply celebrate or demonize their presence in host
economies; he explored both their positive impact (such as local linkages and technology
transfer) and their negative effects (such as crowding out of domestic firms and international
transfer pricing). With the organizational fragmentation that came with global outsourcing and
the rise of more advanced capabilities in the developing world, Lall added questions related to
globalization and technological learning. What is most admirable is that Lall adapted his research
and shifted his policy targets as the world economy evolved, while retaining his central focus on
the key agents of change and their implications for developing countries. This is the path of a
pragmatic, observant and curious mind, guided by a strong moral compass.
The focus of this article is narrower. We assess a single aspect of Lall’s work, his technological
classification of exports, and related research utilizing international trade statistics, from the
point of view of global value chains (GVCs). We see this work on international trade as useful
but ultimately limiting. While the techniques for estimating the technological content of trade
can certainly be further refined by constructing more sophisticated and detailed product-based
analyses of trade flows within or across industries, there is an urgent need to enrich existing
metrics with additional data resources and measures that allow us to investigate GVCs more
directly. In our view, changes in the global economy, and especially the rise of GVCs, have
created measurement problems that require new information and new methods. In an effort to be
constructive as well as critical, we propose one possible approach: the collection of economic
data according to a generic and parsimonious list of business functions.
TRACKING GLOBAL SHIFTS: CONCEPTUAL AND MEASUREMENT ISSUES
Among the enduring mysteries of political economy is why some places surge ahead in the
global economy while others grow more slowly or fall behind in relative or even absolute terms.
Is it sound macroeconomic policy, the development of human capital, protection under the
geopolitical umbrella of a superpower, sector-specific industrial development policies, natural
resource endowments, or some combination that have led to the success of certain countries,
especially in East Asia (Deyo 1987; World Bank 1993)? There are also debates about the optimal
industry structures for technological learning and industrial upgrading. Is a concentrated
industrial structure best because large firms can afford to invest in major research and
development (R&D) efforts, or are open, flexible networks of small and medium-sized firms
35
better able to identify and fit into ephemeral niches of a fast changing global economy (Piore and
Sabel 1982; Amsden 1989; Wade 1990)? The institutional basis for development has also been a
topic of much debate (Evans 1995; Berger and Dore 1996; Hall and Soskice 2001).
For Sanjaya Lall and many others (e.g., Kimura 2007), learning is the key to industrial
upgrading. For places that are behind, learning must, at least in part, come from absorbing
knowledge created elsewhere. Many mechanisms for this have been examined, from arms-length
technological “borrowing” (Amsden 1989) through a range of practices that encompass
technology licensing, reverse engineering, the injection of equipment and know-how through
foreign direct investment, and firm-level adaptation to demands made by both foreign affiliates
and overseas buyers (Gereffi 1994; Feenstra and Hamilton 2006).
Answers to these questions are complex, and debates about what shapes economic development
outcomes will certainly continue. However, we are now at a critical juncture where rising
complexity in the global economy has begun to overwhelm the slow and partial analytical
progress that has been made in the past 25 years. Recent examples, such as how firms based in
the United States, Japan, the Republic of Korea, and Taiwan Province of China interact with
each other and with local firms to produce Apple iPods in southern China for export to world
markets (Linden et al. 2007), illustrate both the intricacies of economic globalization and the
limits of existing data. In this setting some of the core assumptions of mainstream economics—
that demand begets supply, that nations draw mainly on their own knowledge and physical
resources to compete with other nations, that exports reflect the industrial capabilities of the
exporter, that firms and individuals act independently, rationally and at arms-length, and so on—
appear, if not as gross distortions, then as quaint reminders of simpler times. But if the tools of
mainstream economics are being blunted by global integration, so too are those offered by other
social science disciplines, which typically assume levels of institutional and cultural
cohesiveness and economic autarky that no longer exist.
For us, the GVC framework provides a useful guide as we seek answers to questions about the
dynamic political economy of industries.12 GVC analysis highlights three basic characteristics of
any industry: 1) the geography and character of linkages between tasks, or stages, in the chain of
value added activities; 2) how power is distributed and exerted among firms and other actors in
the chain; and 3) the role that institutions play in structuring business relationships and industrial
location. These elements help explain how industries and places evolve, and offer clues about
possible changes in the future. The chain metaphor is purposely simplistic. It focuses on the
location of work and the linkages between tasks as a single product or service makes its way
from conception to end use.
The analysis of GVCs identifies new actors in the global economy (e.g., global buyers and global
suppliers) and shows how their emergence alters the ways that industries are organized and
governed across borders (Gereffi 2005). Recent theorizing about the governance of GVCs
highlights three key determinants that affect the organization and power dynamics within GVCs
(complexity, codifiability, and supplier competence), and characterizes three distinct business
network forms (modular, relational and captive) that lie between the classic duality of arms12
See www.globalvaluechains.org for more detail on this approach and a list of publications and researchers that
directly engage with it.
36
length markets and hierarchies (i.e., vertically integrated firms) (Gereffi et al. 2005). The GVC
governance types were derived from direct field observation in a variety of global industries,
including footwear and apparel (Bair and Gereffi 2001, Gereffi 1999; Schmitz 1999),
horticulture (Dolan and Humphrey 2000), bicycles (Galvin and Morkel 2001), electronics
(Borrus et al. 2000; Lee and Chen 2000; Sturgeon 2002), and motor vehicles (Humphrey 2003;
Sturgeon and Florida 2004).
Qualitative industry research and conceptual theory building of this sort have been extremely
helpful in developing the framework, in identifying emerging trends in GVCs, and in providing
researchers and policymakers with a vocabulary to discuss some of their key features without
getting bogged down in industry-specific nomenclature. The framework has been used,
challenged and extended in recent research on industries such as tourism (Barham et al. 2007),
electronics (Vind and Fold 2007), textiles and apparel (Evgeniev 2008), motor vehicles
(Sturgeon et al. 2008), and coffee and tea (Neilson and Pritchard 2009), and in regions such as
Latin America (Pietrobelli and Rabellotti 2007) and East Asia (Kawakami and Sturgeon
forthcoming).
A major impediment to using qualitative research and conceptual theories to support specific
policy interventions is the lack of comparable and detailed data on the industrial capabilities of
firms, industries, and countries and the roles that they play in the global economy. The GVC
framework provides a conceptual toolbox, but quantitative measures are lacking. While the
development of objective, industry-neutral measures of GVC governance is a laudable goal, and
survey questions are currently being fielded to collect data on the governance character of
interfirm linkages in both cross-border and domestics sourcing relationships, better information
to characterize the roles of firms, regions, and countries in GVCs is urgently needed. 13
In this article, we examine the state of the art in GVC metrics and chart a way forward. First, we
summarize some of the best recent academic research that has used official statistics to examine
issues related to GVCs and industrial upgrading, including Lall’s (2000) technological
classification of exports, Feenstra and Hamilton’s (2007) trade-data archeology, research on
intermediate goods trade, and efforts to enrich trade data by linking it to “microdata” underlying
national statistics and policy programs. We then point to what is perhaps the most glaring data
gap of all: the appallingly poor level of product detail in international services trade.
While the research we review provides useful insights into the dynamics of GVCs and helps to
identify some of the key drivers of industrial upgrading, we are left with a dilemma. The rise in
intermediate goods trade strongly suggests that countries no longer rely only or even primarily
on domestic resources to develop and export products to the rest of the world. Countries and
regions do not make products and deliver services in their entirety, but they have come to
specialize in specific functions within larger regional and GVCs. Surging trade in services
complicates the picture. As a result, industrial output and trade statistics provide a very partial
and even misleading view of where value is created and captured in the global economy.
13
Specifically, Statistics Canada, in an international sourcing survey currently being tested, asks firms if
relationships with important suppliers are simple market relationships or something more complex, and if
transactions involve the exchange of codified or tacit information.
37
Even the best trade statistics, as they currently exist, can only hint at what is happening in GVCs
and how this sort of “integrative trade” (Maule 2006) is shaping development outcomes. If key
GVC-related questions are not asked on any official survey and do not exist on any
administrative form, then existing data resources can never yield adequate results. Thus, there is
an urgent need to collect new information. To illustrate, we present a new business function
classification scheme that is currently being developed and deployed by statistical agencies and
academic researchers in North America and Europe in the hope that it will soon be standardized
and adopted more broadly.14
WHAT TRADE STATISTICS CAN REVEAL ABOUT GLOBAL VALUE CHAINS
Data on international trade in physical goods and commodities are available in considerable
detail online in the United Nations Statistical Division’s Commodity Trade Statistics Database
(known as UN COMTRADE). The database contains import and export statistics reported by the
statistical authorities of nearly 200 countries, from 1962 to the most recent year, currently 2006
to 2008, depending on the country.15 Because these data are collected from many different
national statistical agencies, they vary in quality and coverage. Nevertheless, the UN
COMTRADE database provides information on imports and exports by value and in some cases
by the number of units or volume shipped, according to seven different product (commodity)
lists, the most detailed being the 2002 Harmonized Tariffs Code list, which at the 6-digit level
includes more than 8,000 product descriptions.16
The fine-grained product detail and the ease of access to COMTRADE data have allowed
researchers to create alternatives to the industry classification schemes that its commodity lists
are based on. While industries are an important and often relevant category, they typically
contain products that are very heterogeneous in terms of labor or capital intensity, technological
content, and so on. This section examines three distinct approaches to analyzing trade data that
shed light distinct aspects of GVC development and industrial upgrading. The first is Sanjaya
Lall’s (2000) classification of technological sophistication, which groups products based on their
technological requirements. Increases in “high technology” exports suggest that learning and
industrial upgrading is taking place in the exporting country. Second is the trade-data
archaeology approach developed by Feenstra and Hamilton (2006), which tracks highly detailed
export flows from the Republic of Korea and Taiwan Province of China to the United States over
long periods of time. This approach reveals that specific products, rather than broad industries,
have been key to upgrading in these countries (e.g., microwave ovens from the Republic of
Korea, not white goods in general; computer monitors from Taiwan Province of China, not
electronics in general). Feenstra and Hamilton also tie these exports of narrow product categories
to the strategies of United States retailers and marketers to show how buyer-driven GVCs have
influenced development outcomes in East Asia. The third is work on the relationship between
14
See, for example, the National Science Foundation–funded Project, “A National Survey of Organizations to Study
Globalization, Innovation and Employment.”
http://www.nsf.gov/awardsearch/showAward.do?AwardNumber=0926746&version=noscript.
15
See http://unstats.un.org/unsd/comtrade/.
16
The United States data, published by the Department of Commerce, is available at the 10-digit HTC level and
includes more than 16,000 product descriptions.
38
GVCs and intermediate goods trade. Increases in intermediate goods trade signals the geographic
fragmentation of the production process driven, we argue, by the increasing importance of GVCs
in international trade.
Upgrading As Learning: Sanjaya Lall’s Technological Classification of Exports
Gereffi (2005, p. 171) defines industrial upgrading as “the process by which economic actors—
nations, firms and workers—move from low-value to relatively high-value activities in global
production networks.” Lall et al. (2005) share this view and start with a reasonable assumption,
that the learning required to export high value-added, technology-intensive products will be
greater than for simpler products. Even if the knowledge embedded in imported intermediate
inputs and machinery and know-how from foreign affiliates and global buyers is invisible in
export statistics, as they typically are, we can at least assume that technology-intensive exports
heighten the potential for rapid learning by local actors.
To examine the path of technological learning in the global economy using export statistics, Lall
(2000) devised a technological classification of goods exports. To provide an example of how we
can assess industrial upgrading for export-oriented economies, we examine shifts in the
technology content of China’s and Mexico’s exports over time. Following Lall (2000), we divide
each country’s exports into five product groupings, which are listed in ascending levels of
technological content: primary products; resource-based manufactures; and low-, medium-, and
high-technology manufactures (see Table 1).17 The main contributing industries to each category
(agroforest products, textile and apparel, automotive, and electronics) are broken out to simplify
the analysis.
17
Lall (2000) developed this technological classification of exports based on 3-digit Standard International Trade
Classification (SITC) categories. His article provides the detailed list of products under each category.
39
Table 1. Lall’s Technological Classification of Exports
Classification
Primary products (PP)
Examples
Fresh fruit, meat, rice, cocoa, tea, coffee, wood,
coal, crude petroleum, gas
Manufactured products
Simple RB: Resource based manufactures
Manufac
RB1: Agro/forest based
t-ures
products
RB2: Other resource based
products
LT: Low technology manufactures
LT1: Textile/fashion cluster
LT2: Other low technology
Prepared meats/fruits, beverages, wood products,
vegetable oils
Ore concentrates, petroleum/rubber products,
cement, cut gems, glass
Textile fabrics, clothing, headgear, footwear, leather
manufactures, travel goods
Pottery, simple metal parts/structures, furniture,
jewellery, toys, plastic products
Complex MT: Medium technology manufactures
Manufac
MT1: Automotive products
Passenger vehicles and parts, commercial vehicles,
t-ures
MT2: Medium technology
motorcycles and parts
process industries
Synthetic fibres, chemicals and paints, fertilizers,
plastics, iron, pipes/tubes
MT3: Medium technology
Engines, motors, industrial machinery, pumps,
engineering industries
switchgear, ships, watches
HT: High technology manufactures
HT1: Electronics and electrical
Office/data processing/telecom equip, TVs,
products
transistors, turbines, power gen. eqp.
HT2: Other high technology
Pharmaceuticals, aircraft, optical/measuring
instruments, cameras
Electric current, cinema film, printed matter, special
Other transactions
transactions, gold, works of art, coins, pets
Source: Lall (2000, p. 341).
In Figure 1, panel 1, we see that in 1988, 45 percent of Mexico’s total exports to the United
States market were primary products, the most important of which was oil. In 1993, one year
prior to the establishment of the North America Free Trade Agreement (NAFTA), mediumtechnology manufactures (mainly automotive products) and high-tech manufactures (largely
electronics items) moved ahead of raw materials in Mexico’s export mix. By 2008, over 60
percent of Mexico’s exports of $234 billion to the United States market were in the medium- and
high-technology product categories, followed by primary products with 20 percent of all exports
(which rebounded from their nadir of 10 percent of total exports in 2001) and low-technology
manufactures (such as textiles, apparel, and footwear). Thus, in just two decades, Mexico’s
export structure was transformed from one based on raw materials to one dominated by mediumand high-technology manufactured items.
40
Figure 1. Technological Composition of Mexico’s and China’s Exports
to the United States, 1988 – 2008
Panel 1: Technological Composition of Mexico’s Exports to the United States
50
Primary Products
Resource Based Manufactures
Low Tech Manufactures
Medium Tech Manufactures
High Tech Manufactures
% Export Market
40
30
20
10
0
Total
Exports
US $B
1988
13
1990
16
18
1992
19
37
1994
43
52
1996
66
80
1998
93
2000
2002
2004
2006
2008
102 120 147 136 138 145 167 184 212 223 234
Panel 2: Technological Composition of China’s Exports to the United States
70
60
% Export Market
50
Primary Products
40
Resource Based Manufactures
Low Tech Manufactures
Medium Tech Manufactures
30
High Tech Manufactures
20
10
0
Total
Exports
US $B
1988
3
1990
4
5
1992
6
9
1994
17
21
1996
25
27
1998
33
38
2000
42
52
Source: UN Comtrade (http://comtrade.un.org/db/dqBasicQuery.aspx).
41
2002
54
70
2004
93
2006
2008
125 163 204 233 253
In Figure 1, panel 2, we see the composition of China’s exports to the United States market
during the 1988–2008 period. Unlike Mexico, the leading product category in China’s exports to
the United States market in 1988 was low-technology manufactured goods. These were primarily
made up of a wide variety of light consumer goods, such as apparel, footwear, toys, sporting
goods, housewares. These products accounted for about two-thirds of China’s overall exports to
the United States in the early 1990s. By 2008, however, high-technology exports had increased
to 35 percent of China’s total exports to the United States market, and were virtually tied with
low-technology exports for the top spot in China’s export mix.
Thus, Mexico and China have a number of commonalities in their export trajectories to the
United States market during the past two decades. Both are diversified economies, with a range
of export product types. In both cases, manufactured exports are more important than primary
product or resource-based exports; within manufacturing, high- and medium-technology exports
are displacing low-technology goods. While these export data have limitations as indicators of
industrial upgrading, as we will discuss below, both economies appear to be increasing the
technological sophistication of their exports.
Trade-Data Archaeology
Feenstra and Hamilton (2006) utilize highly disaggregated international trade statistics to shed
new light on the debate surrounding the origins of the “East Asian miracle.” Conventional
explanations of East Asia’s economic success—beginning with Japan in the 1950s and 1960s,
and including the Republic of Korea, Taiwan Province of China, Hong Kong (China), and
Singapore in the 1970s and 1980s—revolve around the role of markets and states in promoting
export-oriented industrialization in this region. The World Bank and neoclassical economists
have favoured the market-friendly explanation, which focuses on the solid macroeconomic
fundamentals in the early East Asian industrializers (World Bank 1993), while other scholars
have highlighted the directive role of the state in promoting this transition (Amsden 1989, Wade
1990, Evans 1995). Feenstra and Hamilton offer a contending demand-side perspective to
account for the sustained export success of the Republic of Korea and Taiwan Province of China,
which ties their performance to the retail revolution and the rise of “big buyers” in the United
States (see also Gereffi [1999]).
Using what they call “trade-data archaeology,” Feenstra and Hamilton recreate the export
trajectories of the Republic of Korea and Taiwan Province of China, not merely at the level of
industries, but by tracing the flow of very specific products over several decades from the early
1970s to the present. This approach reveals that the Republic of Korea’s and Taiwan Province of
China’s dramatic export success was actually concentrated in a handful of product categories,
such as garments, footwear, bicycles, toys, televisions, microwave ovens, computers and office
products. The analysis shows that although exports from Taiwan Province of China and the
Republic of Korea were in the same industries, they specialized in different kinds of products
within these industries: the Republic of Korea’s large vertically integrated chaebol firms
emphasized mass-produced, standardized items, while Taiwan Province of China excelled in
making a wide variety of more specialized products that fit the capabilities of the smaller firms
that dominate the island’s diversified economy.
42
The authors go beyond standard supply-side accounts of East Asia’s export success by showing
precisely how these exports were linked to the “retail revolution” in the United States, where
retailers (such as Sears, JC Penney, Kmart, and Wal-Mart) and companies with global brands
(such as Nike, Liz Claiborne, Disney, and many others) set up international sourcing networks to
tap and expand the global supply base. It was the dynamics within GVCs, as much as any
supply-side market or state-society characteristics, that fuelled the export-oriented development
model that has been promoted by the World Bank and a variety of international development
agencies since the 1980s. The fact that both the Republic of Korea and Taiwan Province of
China developed these “demand-responsive” economies has important theoretical implications
for economic sociology and international trade theories alike (Hamilton and Gereffi 2008).
Examining Intermediate Goods Trade
Merchandise trade has increased dramatically since the 1970s, far surpassing pre-World War I
peaks in most OECD countries. Feenstra (1998) notes a sectoral shift in U.S. imports away from
agricultural products and raw materials and towards capital and technology-intensive goods.
Explanations include trade liberalization, falling transportation costs, and equalization of gross
domestic products (GDPs) among trading countries, given the tendency for countries of similar
size trade more than countries of disproportionate size. Of course, there are many other possible
explanations for these shifts, including rising production skills and better capital stock in poor
countries and speedier transportation, which opens up trade for perishable goods such as fresh
vegetables as well as for goods with very volatile prices, such as computer memory.
The rise of GVCs is not only enabled by these factors, but is itself a cause of trade increases. As
Feenstra (1998, p. 36) argues, the geographic fragmentation of production causes increases in the
volume of total trade because intermediate inputs may cross borders several times before final
products are delivered to end users. Thus, the trade content of an average product rises when it is
made in the context of GVCs.
The fact that intermediate goods trade is rising much faster than overall trade has stimulated a
vast body of research and multiple labels, including a new international division of labor (Fröbel
et al. 1980), multistage production (Dixit and Grossman 1982), slicing up the value chain
(Krugman 1995), the disintegration of production (Feenstra 1998), fragmentation (Arndt and
Kierzkowski 2001), vertical specialization (Hummels et al. 2001), global production sharing
(Yeats 2001), offshore outsourcing (Doh 2005), and integrative trade (Maule 2006). Sturgeon
and Memedovic (forthcoming), using the UN’s broad economic categories of consumption,
capital, and intermediate goods, calculate that global trade in intermediate goods has far
outpaced these other categories (Figure 2). This rise is most dramatic after 1992, when the
developing world was linked more systematically in GVCs. The share of worldwide imports of
intermediate goods by developing countries increased from only 25.5 percent in 1992 to 35.2
percent in 2006. During this period total trade in intermediate goods grew 2.2 times in
industrialized countries and 3.4 times in developing countries.
43
Figure 2. Intermediate, capital, and final goods trade, 1962–2006
(millions of constant U.S. dollars)
6,000
5,000
4,000
Capital goods
Consumption goods
Intermediate goods
3,000
2,000
1,000
0
1962
1967
1972
1977
1982
1987
1992
1997
2002
Source: Sturgeon and Memedovic (forthcoming).
While soaring intermediate goods trade is a strong indicator of the rise of GVCs, their growing
dominance of world trade can lead to odd and confusing metrics. For example, because Malaysia
imported so many intermediate goods for inclusion in exports, its ratio of exported goods and
services to GDP in 2005 reached 123.4 percent (World Development Indicators 2007). Such
ratios are not uncommon in classic entrepôt economies such as Singapore and Hong Kong
(China), and as a comparative measure of trade integration this is fine, but upon seeing such
statistics without reference to GVCs, one has to wonder how a country can export more than it
produces.
Clearly, the global economy is changing. Rising intermediate goods trade means that goods are
flowing, increasingly, within the same industry. Research on intraindustry trade (Grubel and
Lloyd 1975; Lloyd and Lee 2002) has shown steady increases of about 4–5 percent per year in
countries trading the same or seemingly similar products. This challenged the central tenet of
Ricardian trade theory: country specialization according to factor-based comparative advantage
that would lead only to interindustry trade. Finger (1975) claims that coarse industry
classifications disguised vast heterogeneity within industries; in other words, countries could
specialize within the same industry, especially in intermediate inputs versus final goods.
For Krugman (1991), intraindustry trade is driven by firms seeking increasing returns from largescale production, thereby generating exports, while consumer demand for product variety
stimulates imports of very similar products. Although this work was based on horizontal
differentiation (of similar products), the quality ladder-growth models from Grossman-Helpman
44
(1991), which are formally very similar to Krugman’s model, have a vertical dimension that
includes intermediate goods. Others have tested and refined these theories in the context of East
Asia’s economic rise (Ng and Yeats 1999) and provided evidence of increasing “vertical”
specialization in intermediate inputs (Hummels et al. 2001). Using updated statistics, Brülhart
argues that “. . . since the 1990s, [the increase in intraindustry trade] appears to be driven to a
significant extent by the international fragmentation of vertical production chains” (Brülhart
2008, abstract).
Our argument is that trade statistics can only hint at the changes occurring in the global
economy. Trade statistics alone contain very partial information about the location of value
added, and no information about ownership of productive assets and output, where profits are
reaped, or how these increasingly complex systems are coordinated. Certainly, work will
continue on the causes and meaning of interindustry trade. But there are limits to what can be
learned from trade statistics alone.
USING ADMINISTRATIVE AND MICRODATA RESOURCES TO UNDERSTAND
GLOBAL VALUE CHAINS
Linking trade statistics to other datasets can enhance their usefulness. Through careful matching,
or by taking advantage of especially rich administrative data,18 researchers can sometimes push
beyond the limitations of published statistics. A host of government programs collect detailed
economic data. Typically more detailed microdata underlie what is ultimately made available to
the public. While these data are usually confidential, researchers who gain security clearance and
have their proposals accepted by data collection agencies can gain access, as along as
government personnel screen the results before the research is published. Some microdata sets
have also been assembled by data agencies and released, with confidential information removed,
as public-use files. Over the past decade, a burgeoning body of research has relied on
government-collected microdata. In this section, we provide a few examples.
Feenstra and Hanson (2004, 2005) take advantage of administrative data from Mainland China
and Hong Kong (China) to reveal new information about the workings of GVCs. Specifically,
the data contain reexport values for Hong Kong and information about factory and input
ownership in China. These data allow the authors to estimate the mark-up charged by Hong
Kong–based GVC “intermediaries” such as Li and Fung, a trading company. The authors also
use these data to calculate the share of China’s exports to Hong Kong that are reexported (45.4
percent in 1998), an indicator of the important coordination role that companies like Li and Fung
play in GVCs, especially in apparel and other consumer-goods industries. By taking advantage
of data that describe the ownership of factories exporting from China, the authors are able to
show that independent suppliers working under “export processing” arrangements (i.e., suppliers
that are provided with inputs by intermediaries and their customers) are much more likely to
send goods through Hong Kong for reexport than exporting factories that are wholly owned by
non-Chinese firms.
18
Governments collect data for the purpose of administering their programs such as tax collection, compliance with
environmental protection laws, and the like. For this reason, such data are typically referred to as “administrative
data.”
45
Feenstra and Spencer (2005) use the same Chinese data, from 1998 through 2000, to explore the
relationship between outsourcing arrangements (arms-length vs. contractual) and the proximity
of suppliers (on-shore vs. offshore). They find that relationships vary according to the
technological sophistication of the product being outsourced. The more technologically
sophisticated the product, the more likely it is that firms will source from affiliates or outsource
to suppliers located nearby. Dani Rodrick and his collaborators (Haussman et al. 2006) use these
data to show that the basket of goods exported by China is of higher technological content than
would be predicted by its GDP per capita (using averages for all other countries’ export mixes).
By linking these same data to Chinese input-output data, Dean, Fung, and Wang (2007) estimate
that China’s “vertical specialization,” that is, the use of imported intermediate inputs in exported
goods, increased between 1997 and 2002 in most industries. This is the opposite of what one
would expect. Instead of engaging in progressive import substitution as domestic capabilities
rise, as most theories of development predict, China increased its reliance on imported
intermediates as exports increased. Here we see that, because of the intricacies of production and
trade networks within GVCs, we cannot assume deterministic causal linkages between export-led
industrialization, the technological content of exports, and industrial upgrading.
Researchers have creatively used microdata to explore specific questions related to GVCs. For
example, Bernard et al. (2005) link administrative data from U.S. census mailing lists to the universe
of import and export transactions for 1993–2000, revealing a detailed picture of the characteristics of
firms that do and do not trade. 19 Harrison and McMillan (2006) and others have used the parent and
foreign affiliate microdata from the Bureau of Economic Analysis surveys on TNCs to examine the
relationship between affiliate activity and United States employment. Swenson (2005) has examined
the permanency of offshore assembly arrangements using extremely detailed data from United States
International Trade Commission (USITC) reports. Kletzer (2002) has used microdata from the
Displaced Worker Survey to explore the experiences of workers displaced from manufacturing
industries associated with increased foreign competition, and has made policy recommendations
based on her findings.
Administrative microdata from public surveys and linked datasets can enrich our view of how
domestic firms engage with the global economy. Microdata collected from TNCs, for example,
when combined with data on international trade, can provide new information about the crossborder activities of TNCs and how they use local resources in offshore locations. Such
approaches can be difficult to replicate and extend, however, because not all researchers can
access confidential microdata, and because the painstaking work of cleaning and matching raw
microdata files can be very difficult for other researchers to understand and replicate.
Furthermore, unique administrative datasets tend to be available only for individual countries,
and data collected in support of specific policy initiatives are commonly phased out after the
19
We are referring here to the United States Census Bureau’s Business Register, which is the sampling frame used
for the Economic Census. Data included are business name, address, a unique establishment-level identifier,
industry, employment, and the identity of the firm that owns the enterprise. Data about ownership allows the
enterprises in the Business Register to be aggregated to the firm level. Jarmin and Miranda (2002) have assembled
the Business Register into a time-series for 1976-2002, referred to as the Longitudinal Business Database (LBD).
46
programmes they were intended to support come to an end. As a result, studies based on
microdata can have limited scope with regard to multiple countries and longer-term trends.
WHAT TRADE STATISTICS HIDE
The easy availability and richness of UN COMTRADE data has led to their wide use among
researchers and policymakers. However, we need to keep in mind what trade statistics do not tell
us, and even what they might obscure. First, trade data contain no actual information about the
process by which products are made. Certain production processes, such as semiconductor wafer
fabrication, involve the manipulation of items so small, or require tolerances so exact, that they
have moved beyond the limits of human dexterity and must always be carried out by machines.
Other processes, such as sewing, have so far resisted automation and can only be done by hand.
But for a very wide range of products and processes, the labor content of production is variable.
The degree of labor or capital intensity used in production is, in many instances, a strategic
managerial choice rather than an intrinsic characteristic of the product. Thus, we cannot rigidly
associate technological content or capital requirements with most specific categories or classes of
products. Industries are even poorer indicators of technological sophistication.
Furthermore, the technological content of high-technology exports may be embodied in imported
components, subsystems, or production equipment. The highest value-added elements of hightechnology exports from developing countries are often produced in a third country. Even if
these “high-tech” inputs are produced locally, and final assembly processes are truly technology
intensive, they may be carried out by foreign-owned and operated firms with few meaningful
linkages to the local economy. With rising wages, worker militancy, political friction, or even a
prolonged natural disaster, such footloose firms might easily pack up and move elsewhere. Thus,
trade statistics run a real risk of overstating the technological competence of exporters, and
especially of local firms.
Even when production is carried out by local firms and is truly technology intensive, the reality
of GVCs is that the innovative work of product conception, design, marketing and supply-chain
management may well continue to be conducted outside of the exporting country. These
“intangible assets” cannot be measured by current international trade statistics. The value of
imports plus the intangible assets held by the most powerful firms in GVCs, such as lead firms
with global brands, suppliers with platform leadership (Gawer and Cusumano 2002), and large
retailers, can be extremely high.
For example, Linden et al. (2007) estimate that only $4 of the $299 retail price of an Apple 30
gigabyte video iPod MP3 player is captured in China, where they are assembled and tested by
the contract manufacturers based in Taiwan Province of China, Hon Hai (also known as
Foxconn), Asustek, and Inventec. This is in part because iPods are assembled from components
made mostly in other countries, such as the United States, Japan, and the Republic of Korea. But
more importantly, it is because Apple, which conducts high-level design work and software
development in house and orchestrates the product’s development, production, marketing, and
distribution, is estimated to capture $80 of the sale price. This study also estimates that $83 is
captured in the United States by Apple’s technology suppliers and by retailers. Clearly, assigning
47
the $183 per unit wholesale price of exported iPods (as would be reported in trade statistics) to
the Chinese economy misrepresents where value is created in the global economy. It also would
be a mistake to conclude that Chinese firms have the capability to develop and market products
such as the iPod simply because the country is the source of exports.
A Glaring Data Gap: Services Trade
The easy availability and richness of UN COMTRADE data has tilted research on international
trade toward the goods sector. While this work has contributed greatly to our understanding of
international trade and its impacts on various national economies and industries, the lack of
similar detail or global coverage on international trade in services has created a significant
knowledge gap. In the case of the United States, the Bureau of Economic Analysis collects
import and export data for only 17 service product categories (see Table 2). Statistics Canada
collects only 28, and the OECD, which relies on member countries for data, publishes only 11.
Contrast the poor detail in traded services with detail on goods in the COMTRADE database
(8,000 product codes) and the magnitude of the data gap becomes clear.
Because of this data gap, we lack the basic knowledge about services trade needed to even
glimpse trends in industrial upgrading driven by services. The paucity of detail in services means
that we have no information about what is happening in the service product categories that have
been mentioned as moving offshore from developed to developing countries, including backoffice functions such as accounting, customer support, R&D, and software programming.
Why are the data resources related to services so poor? One reason is that the data are difficult to
collect. While companies might track the source of every physical input to manufacturing, for
warranty or quality control purposes, services expenditures are typically grouped into very
coarse categories, such as “purchased services.” The absence of tariffs on services, and their
nonphysical character, means that when service work moves across borders, no customs forms
are filled out and no such data are generated. Another reason is that service work has historically
been thought to consist of nonroutine activities that require face-to-face contact between
producers and users. Services as different as haircuts and legal advice have traditionally been
consumed, in place, as soon as they are produced. The customized and ephemeral nature of many
services has led them to be considered “nontradable” by economists or at least very “sticky” in a
geographic sense relative to the production of tangible goods. Finally, services have long been
viewed as ancillary to manufacturing, either as direct inputs (e.g., transportation) or as services
provided to people who worked in manufacturing (e.g., residential construction, retail sales, etc.).
As such, services have been viewed as a by-product, not a source, of economic growth. Thus,
data collection on services has been given a low priority by statistical agencies.
48
Table 2. The 17 product categories collected by the U.S. Bureau of Economic Analysis
for traded services
Travel, passenger fares, and
other transportation services
(1)
Royalties and license fees (2)
Insurance services (5)
Financial services (4)
Business, professional, and technical services
Computer and information
Management and consulting
services
services (9)
Computer and data processing
services (7)
Database and other information
services (8)
Construction, architectural,
Industrial engineering services (12)
engineering services (11)
Installation, maintenance, and
Advertising services (15)
equipment repair services (14)
Other business, professional,
and technical services (17)
Source: U.S. Bureau of Economic Analysis.
Education (3)
Telecommunications services
(6)
Research, development and
testing services (10)
Operational leasing services (13)
Legal services (16)
Nevertheless, services trade is burgeoning, both domestically and internationally.
Computerization is allowing a growing range of service tasks to be standardized, fragmented,
codified, modularized, and more readily and cheaply transported between producers and
consumers who might be at great distance. As in goods production, the application of
information technology to the provision of services allows some degree of customization within
the rubric of high volume production, or what Pine and Davis (1999) call “mass customization.”
With computerization and inexpensive data storage, the second defining feature of services, that
they cannot be stored, has also become less true than in the past. With deregulation, business
process outsoucing, and the rise of the Internet, services have become the focus of intense
international competition and rampant innovation. Clearly, the assumptions behind current data
regimes have changed and statistical systems must catch up.
Recent progress has been made in the context of NAFTA. In the spring of 2006, the U.S. Census
Bureau, in collaboration with its counterpart agencies in Canada and Mexico, completed the
development of 99 detailed product lists that identify and define the significant products of about
370 service industries. Work to date on the North American Product Code System (NAPCS) has
focused on the products made by service industries in 12 two-digit industry sectors (48–49
through 81). In all, more than 3,500 individual service products have been defined. The NAPCS
product definitions are extremely detailed in terms of what they do, and in many cases do not,
include. This level of detail, if fully deployed, would go a long way toward filling the data gap in
services trade.20
20
For more information on NAPCS, see http://www.census.gov/eos/www/napcs/napcs.htm.
49
To sum up, data resources are falling behind economic realities. Innovative work to create new
classification schemes from disaggregated datasets, to mine microdata from government surveys
and administrative records (as well as from private sources), and to combine and match data to
create new data resources, is breaking new ground and providing important insights. A few of
the most severe data gaps could eventually be filled. However, more needs to be done to collect
data specifically designed to provide insights into the characteristics and effects of GVCs. Work
of this sort is proceeding along multiple fronts, including the surveys that test the GVC
governance framework developed by Gereffi et al. (2005) and the quantification of value capture
in specific GVCs (Linden at al. 2007). Equally important is the ongoing stream of detailed fieldbased research on the functioning of GVCs, in particular industries and places (e.g., Kawakami
and Sturgeon forthcoming). In the next section, we propose another approach: the collection of a
broad range of economic data, such as employment, sourcing locations, and job characteristics
according to an exclusive, exhaustive, parsimonious, and generic list of business functions.
COLLECTING NEW DATA ON BUSINESS FUNCTIONS
Vertical fragmentation and the growth of integrative trade—the very stuff of GVCs—has served
to expand the arena of competition beyond final products to the vertical business function slices
that can be offered (horizontally, to diverse customers) as generic goods and services within and
across industries. This dynamic has raised the performance requirements for firms and workers
that may have been insulated from global competition in the past. Workers, almost regardless of
their role, can suddenly find themselves in competition with a range of consultants, vendors,
suppliers, contractors, and affiliates from places both far and near. Global value chains raise,
among other things, the possibility that entire societies can become highly specialized in specific
sets of business functions, while others fail to develop or atrophy. Development paths that
include heavy GVC engagement can have positive or negative consequences for wealth creation,
employment, innovation, firm autonomy, social welfare, and economic development (Whittaker
et al. forthcoming). Despite their growing importance as discrete realms of value creation,
competition and industry evolution, we currently have no standard method for collecting data
about business functions.
While there are a host of business functions that have long been disembodied from specific
industries (e.g., from janitorial to IT to manufacturing services), qualitative research has shown
that managers often experiment with a wide variety of “make” or “buy” choices and on- or
offshore sourcing (Berger et al. 2005). Decisions about how to bundle and unbundle, combine
and recombine, and locate and relocate business functions have become a central preoccupation
of strategic decision-making. Because industry classification schemes typically describe only the
main output or process of the firm, and input-output statistics refer only to those products the
firm buys or sells, existing enterprise and establishment-level data resources are not well suited
to capturing the dynamics of business function bundling or revealing the spatial and
organizational patterns that result.
In our view, this data gap will become more important over time as the capabilities that reside in
the domestic and global supply-bases continue to rise, increasing the potential for fragmenting,
outsourcing and relocating a wide variety of business functions. A standardized list of exclusive
50
and generic business functions is needed. An exclusive list will have no overlap between
categories and will account for all of the functions of the firm. A generic list will be equally
applicable to all firms and organizations, regardless of industry. The list should be extremely
parsimonious at first, with detail collected only after the main categories have stabilized through
field testing. While this is a difficult and time-consuming prospect, work to develop business
function lists and deploy them in surveys is well under way.
Developing, Deploying, and Refining Business Function Lists: A Brief History
To our knowledge, the earliest use of a business function list to collect economic data was for the
EMERGENCE Project (Huws and Dahlman 2004), funded by the European Commission. This
research uses a less-than-generic list of seven business functions tailored to collect information
about the outsourcing of information technology–related functions, such as software
development and data processing. Industry-specific bias in business function lists can simplify
data collection and focus research on specific questions, but the results cannot be easily
compared to or aggregated with other data, and they increase the risk of creating nonexhaustive
lists. When business function lists are nonexhaustive, they leave some functions unexamined and
block our view of how specific business functions contribute to the total employment or output
of a firm. Business function lists should seek to include the full range of activities that all
establishments must either do in house or have done by others, regardless of industry.
In his 1985 book, Competitive Advantage, Michael Porter publishes a list of nine generic
business functions: R&D, design, production, marketing and sales, distribution, customer
service, firm infrastructure, human resources, and technology development. A list similar to
Porter’s was developed for the European Union (EU) Survey on International Sourcing (Neilsen
2008) and adopted by Statistics Canada for the Survey of Changing Business Practices in the
Global Economy. This list, while not industry-specific in any way, was not fully exhaustive
because it included an “other functions” category. Such categories are useful as checks on the
exhaustiveness of the list used, but researchers should then combine them with an existing
category or, if needed, define a new, exclusive category, rather than accepting an undefined
category of data.
Firms, especially at the establishment level, typically have a main output, be it a product or
service. The main operational function that produces this output is associated with the firm’s
standardized industrial code. Instead of counting all output and employment under this
classification, as business censuses typically do, business function lists can be used to measure
economic activity (e.g., employment, occupational mix, wages paid, etc.) in other functions as
well. In business function frameworks, this main productive function has been designated
variously as “production” (Porter 1985), the “core function” (Neilsen 2008), and “operations”
(Brown 2008). In contrast, the EMRGENCE project list (Huws and Dahlman 2004) and a more
recent list developed by the Offshoring Research Network for the purpose of detecting R&D
offshoring (Lewin et al. 2009) did not include a category for the firm’s main operational
function. Instead it used a list of commonly outsourced functions (product development, IT
services, back office functions, call centers, etc.). A business function list cannot be considered
exhaustive unless it includes a category that captures the main productive function of the firm, a
function that can be partially or even completely outsourced.
51
The Bureau of Labor Statistics’ (BLS) Mass Layoff Statistics (MLS) program has developed a
list to collect data on business functions fulfilled by workers who have been separated in largescale layoffs in the United States (Brown 2008). In the 2007 MLS survey of establishments,
respondents were asked a question about the primary and secondary roles, or “business
functions,” performed by laid-off workers. According to Brown (2008, p. 56), “ ‘Do not know’
responses to the business function question remained low [less than 6 percent], indicating that
the correct person is being reached for the interview and that most respondents in fact think in
terms of business functions.” In other words, the BLS found business function data to be highly
collectable because company officials appear to recognize the business function concept. A
tabulation of respondents’ literal responses generated a very long, nonexclusive list of business
functions that were then coded by BLS personnel to create detailed, mutually exclusive
categories. This list was further coded to nine higher-level business functions (named “business
processes” in the MLS) similar to the Porter list. It is the bottom up methodology used by the
BLS—beginning with literal responses rather than using a list that researchers develop
subjectively or iteratively with industry informants—along with its exhaustive, exclusive, and
generic character, which gives us a high level of confidence in the BLS list.
A Proposed List of Business Functions
The growing use of business function lists in survey research suggests a need to delve within the
firm to observe the details of organizational design, organizational change, outsourcing and
industrial location. Clearly, new realities are spurring researchers to develop these new metrics.
In our view, the sooner a business function classification scheme can be standardized and
broadly deployed the better.
Table 3 presents a proposed list of 12 business functions, along with their definitions. The list
adds four business functions to the 2007 BLS MLS list. First, there is a function called “strategic
management.” This reflects the common separation of the command, control, and strategysetting activities of top management from more mundane managerial functions that can
sometimes be located offshore and/or carried out in supplier firms. The most recent BLS MLS
surveys distinguish strategic management from a set of “general management” functions.
Second, because they typically occur at nearly opposite ends of the value chain, procurement has
been separated from distribution, transportation, and logistics. Third, our list breaks out
“intermediate input and materials production” from operations. This is meant to capture the very
common practice of externally sourcing physical parts or blocks of services for inclusion in
larger products and systems. In the BLS MLS list intermediate input production is considered
part of operations. Fourth, because they contain very different activities, firm infrastructure has
been broken out from general management (and corporate governance). Despite these
differences, the lists are compatible since the functions in Table 3 can be combined to match the
BLS MLS list.
52
Table 3. Twelve generic business functions and their definitions
Business function
1) Strategic management
2) Product or service
development
3) Marketing, sales, and
account management
4) Intermediate input and
materials production
5) Procurement
6) Operations (industry code)
7) Transportation, logistics,
and distribution
8) General management and
corporate governance
9) Human resource
management
Definitions
Activities that support the setting of product strategy (i.e., deciding what "new
product development" works on), choosing when and where to make new
investments and acquisitions, or sales of parts of the business, and choosing
key business partners (e.g., suppliers and service providers).
Activities associated with bringing a new product or service to market,
including research, marketing analysis, design, and engineering.
Activities to inform buyers including promotion, advertising, telemarketing,
selling, retail management.
The fabrication or transformation of materials and codification of information
to render them suitable for use in operations.
Activities associated with choosing and acquiring purchased inputs.
Activities that transform inputs into final outputs, either goods or services.
This includes the detailed management of such operations. (In most cases,
operations will equate with the industry code of the establishment or the
activity most directly associated with the industry code.)
Activities associated with transporting and storing inputs, and storing and
transporting finished products to customers.
Activities associated with the administration of the organization, including
legal, finance, public affairs, government relations, accounting, and general
management.
Activities associated with the recruiting, hiring, training, compensating, and
dismissing personnel.
10) Technology and process
Activities related to maintenance, automation, design/redesign of equipment,
development
hardware, software, procedures, and technical knowledge.
11) Firm infrastructure (e.g.,
building maintenance. and
Activities related to building maintenance, and ITC systems.
IT systems)
12) Customer and after-sales
Support services to customers after purchase of good or service, including
service
training, help desks, customer support for guarantees, and warranties.
Source: Adapted by the authors from Bureau of Labor Statistics, Mass Layoff Statistics Program.
Collecting Data on the Geography of Business Functions
Although business function data can be used to inform other research questions, as the BLS’
MLS program does in identifying the functional role of laid-off workers, our main interest in
using it to identify patterns of business function bundling (i.e., organizational design), and the
locational characteristics of outsourcing and offshoring. Because business functions can be
bundled and located differently, we can identify four nonexclusive quadrants for any given
function: 1) domestic in house, 2) domestic outsourced, 3) offshore in house (i.e., the MNC
affiliate), and 4) offshore outsourced. However, it is important that business function surveys that
seek to capture data on global engagement are designed, not only to capture all four, but also the
ways that firms combine them. Firms can, and typically do, combine internal and external
sourcing of specific business functions. For example, some intermediate inputs may be produced
in house while others are outsourced. Operations may be outsourced, but only when internal
capacity is fully utilized. Firms might combine internal and external sourcing for strategic
reasons (Bradach and Eccles 1989).
53
The same can be said of location. Managers can decide to locate business functions in proximate
or distant locations, in high- or low-cost locations, near customers, suppliers, specialized labor
markets, and so on, but most typically they combine these approaches and motives. This is why
detailed information about the location of business functions is of great interest. Surveys that
identify sourcing locations and either domestic or international are not very helpful. Outsourcing
from the United States to Germany, for example, will likely involve different functions and have
very different and motivations and implications than outsourcing from the United States to
China. But even on the domestic front, outsourcing to a vendor in the same city is very different
from outsourcing to a supplier located in a distant, rural location.
The surveys on international sourcing fielded by Eurostat, Statistics Canada, and the Offshoring
Research Network collect no data on domestic locations and use predetermined lists of
geographic locations to identify countries of great interest (e.g., India and China), but combine
others into vast, amorphous groupings (e.g., “other Asia”). It is better, in our view, to ask
respondents to provide geographic information according to city and country. In this way, a
single question can begin to identify, with great precision, both domestic and international
patterns of outsourcing and offshoring. Geographic aggregations can be made after the fact, and
detailed locational coordinates can allow the use of geographic information system software to
create and examine a host of potentially important variables (e.g., clustering, distances, travel
times, prevailing labor market conditions).
Data collected according to business function can provide researchers and policymakers with a
rough map of the value chain; reveal the roles that domestic establishments, firms, and industries
play within GVCs; and offer a unique view of the competitive pressures facing domestic firms
and industries. Over time, it will be possible to develop a hierarchy of business functions to
provide information about business functions in greater detail, but in the shorter term a
parsimonious, high-level list can provide important information, such as an at-a-glance
perspective on how enterprises bundle value chain functions and a benchmark for how this is
changing. As metrics for the key variables of GVC governance and the five GVC governance
modes described earlier are developed, they can be used to characterize the internal and external
linkages between specific business functions, testing our assumptions about the relationships
between GVC governance and the “offshorability” and location of work. Nationally
representative surveys can begin to characterize business function gaps and specializations in
specific countries, while international surveys can develop comparisons between trading
partners. When combined with existing data on employment, occupations, wages, worker career
paths, firm performance, E-commerce, trade, etc., new data on business functions will open up
important new avenues for research and policy analysis.
A New European Survey on Business Functions
To provide an example of the usefulness of business function data, we present some preliminary
data from the EU Survey on International Sourcing. So far, the survey has been administered in
14 out of 27 EU member states, and 60,000 responses have been collected, but only the data
from four Nordic countries have been tabulated (see Nielsen [2008] for details). Figure 3 and
Table 4 and Table 5 show the results from Denmark, where the survey was carried out as a
54
census for all 3,170 private sector nonagricultural enterprises with 50 or more employees.21
Because a few of the core questions were mandatory, the response rate for this group of
establishments was 97 percent. The questions about business functions on this survey were
straightforward: Were business functions outsourced domestically or internationally in the 2001–
2006 period (Table 4), and if so, what kind of business partner was used (Table 5), and (from a
predetermined list) where were internationally sourced functions located (Figure 3)?
The data in Table 4 show that Danish firms sourced the majority of business functions in house.
About 88 percent were not engaged in international sourcing of any kind. Facilities management
was the most commonly outsourced function (37 percent), but because vendors provide these
services on site, the source was invariably domestic. The business function that was sourced
internationally the most frequently was the “core” function (10 percent of all firms), analogous to
“operations” in Table 3, followed by information technology and communications (ITC)
services. Twenty-nine percent of the 1,567 functions reported as internationally sourced were
core functions, followed by ITC services (16 percent), distribution and logistics functions (13
percent), engineering functions (11 percent), administrative functions (10 percent), marketing
and sales functions (10 percent), and research and development functions (9 percent).
These data support anecdotal evidence that international sourcing is most advanced in
manufacturing (a “core” function for goods producing firms). This assumption gains further
support when firms reporting their core function as manufacturing are compared to serviceproducing firms. Only 28 percent of service producing firms in Denmark reported international
sourcing of their core function, while 70 percent of manufacturing firms did so (Nielsen 2008, p.
24). Table 5 shows that less than half of the reported international sourcing by Danish firms in
the 2001–2006 period was to independent firms. The bulk of in-house international sourcing
went to existing affiliates, as opposed to recently acquired or newly established “greenfield”
affiliates.
Table 4. External and international sourcing of business functions by Danish firms, 2001–2006
Business Function
Core function
ICT services
Distribution and logistics
Administrative functions
Engineering
Marketing, sales etc.
R&D
Other functions
Facility management
Domestically
outsourced
Not outsourced
88%
71%
82%
90%
88%
91%
94%
96%
63%
4%
24%
15%
7%
9%
6%
3%
4%
37%
Internationally
sourced
10%
6%
4%
4%
4%
3%
3%
1%
0%
Source: Eurostat International Sourcing Survey, courtesy of Statistics Denmark (Nielsen 2008).
Notes: n=3,170 Danish enterprises with more than 50 employees. Rows may not add to 100 percent because a few
firms reported more than one source for a given business function.
21
The survey was also administered to 1,968 smaller Danish manufacturing and business services firms. For
simplicity’s sake, these data are not presented in this paper. In general, they show similar patterns but slightly less
domestic and international outsourcing across business functions than the sample of larger firms.
55
Table 5. Internationally sourced business functions by Danish firms, by supplier type, 2001–2006
Business Function
Core function
Distribution and logistics
Marketing, sales etc.
ICT services
Administrative functions
Engineering
R&D
Facility management
Other functions
Existing
affiliate
Recently
acquired
affiliate
Recent
greenfield
affiliate
Independent
firm (< than
50% owned)
29%
43%
48%
46%
50%
33%
34%
NA
9%
8%
5%
8%
3%
3%
6%
8%
NA
9%
18%
15%
14%
6%
13%
16%
9%
NA
0%
46%
37%
30%
44%
34%
45%
49%
NA
81%
Source: Eurostat International Sourcing Survey, courtesy of Statistics Denmark (Nielsen 2008).
Notes: n= 611 Danish enterprises engaged international sourcing.
Figure 3 summarizes the geography of international sourcing by Danish firms. It shows that new
European Union (EU) member states (mostly in Eastern Europe) account for 31 percent of the
cases of international sourcing of core functions during the 2001–2006 period, followed by
China (22 percent) and old EU member states (19 percent). When the focus is shifted to ITC
services, the importance of the new member states falls to only 8 percent, while old member
states account for 57 percent of the cases of international sourcing. India, a country typically
identified as a destination for ITC outsourcing in the popular press and in qualitative research, is
identified as a source country in 12 percent of the cases of ITC sourcing, in comparison with
only 5 percent of the international sourcing cases for core functions. International outsourcing of
R&D and engineering functions is also concentrated in Western Europe (42 percent), with China
(9 percent) and “other Asian” countries (8 percent) playing a larger role than in ITC services.
Interestingly, the role of India in R&D outsourcing is very small. The combined shares of
marketing, distribution, and administrative functions show a more balanced pattern across
locations.
The results presented here are largely unsurprising. They confirm both qualitative GVC research,
and to some extent popular perceptions. Of the business functions that are sourced outside of
Denmark, 30–50 percent are outsourced to independent suppliers, a substantial but not dominant
share. Existing affiliates provide most of the in-house international sourcing, but international
acquisitions and the establishment of new “greenfield” facilities are not unheard of. Core
functions, mostly manufacturing, are most commonly outsourced and offshored, followed by
ITC services. Functions based on tacit and local knowledge, such as marketing and sales,
engineering, and R&D are less likely to be internationally outsourced or offshored. Most
international sourcing by Danish firms is within Europe, but China is a popular location for
sourcing core functions (mainly manufacturing). While India is more likely to be a source
location for ITC service functions (12 percent of cases) than for core functions (5 percent of
cases), it is notable that the majority (57 percent) of instances of international ITC services
sourcing are to the original 12 member states of the European Union.
56
While it is important to have our impressions confirmed, the greater value of these data is that
they establish a baseline for future research. Is the practice of outsourcing to independent
suppliers becoming more prevalent? Will India grow as a location for ITC sourcing at the
expense of old European Union member states? Will the outsourcing of engineering and R&D
functions grow, and if so, where? Will service-producing firms increase the outsourcing and
offshoring of core functions [operations]? If these are trends, then how quickly will they
progress? Will Eastern Europe lose out to East Asia? Such questions comprise some of the most
pressing policy questions of the day. When and if new rounds of business function data are
collected, we will be in a much better position to provide answers.
What the Eurostat international sourcing survey did not collect was employment and wage data
according to business function. Such data would begin to quantify the importance of specific
business functions within firms, industries, and countries, and provide a benchmark for
comparison with other countries that could reveal patterns of organizational design and national
specialization within GVCs. It is our hope that future surveys will collect these data. One way
could be to code census data that reveals performance metrics such as sales, employment and
payroll according to a business function framework.
57
Figure 3. International sourcing of business function by Danish firms, 2001–2006
Core function
ITC services
USA and
India
USA and
Canada
4%
Canada Other*
2%
5%
India
12%
EU12 (new
Other European
7%
member states)
EU12 (new
member
Other* states) 8%
1%
China
4%
31%
Other
European
9%
Other Asian
10%
Other Asian
5%
EU15 (old
member states)
China
19%
EU15 (old
member
states)
57%
22%
Marketing, distribution, and
administrative functions
R&D and engineering functions
USA and
Canada
Other*
Other*
3%
EU12 (new
USA and
member
India
Canada
states) 18%
5%
2%
EU12 (new
member
states) 11%
Other European
6%
China
India
9%
Other Asian
8%
China
5%
12%
Other European
9%
Other Asian
9%
EU15 (old
member states)
EU15 (old
member states)
30%
46%
Source: Eurostat International Sourcing Survey, courtesy of Statistics Denmark (Nielson 2008).
Notes: Other is Latin and South America plus Africa. Other Europe is Switzerland, Norway, Turkey, Russia, Belo
Russia, Ukraine, and the Balkan states. n=611 Danish enterprises engaged international sourcing.
58
CONCLUSIONS
In the mosaic of value chain specialization and intermediate goods flows that underlie the most
recent trends in global integration, ownership and capability development cannot be so easily be
linked to the domestic context, even if we allow that it is based in part on “borrowed”
technology. The implications for policy are far reaching. How can workers, firms, and industries
be provided with the best environment for engaging with the global economy? How can we be
sure that enough wealth, employment, and innovative capacity are generated at home as global
integration proceeds? How much national specialization—and by extension, interdependence
with other societies—is too much? These are open questions. Even if policymakers seek few
direct interventions in the areas of trade, industrial, or innovation policy, global integration can
make the process of economic adjustment more difficult because it accelerates the pace of
change.
Because the picture of global integration provided by current official statistics is incomplete, the
causal links to economic welfare indicators such as employment and wages tend be weak and
unconvincing. New thinking is required to develop useful insights into the character and
implications of our increasingly globally integrated national economies. Perhaps the most
pressing need is for new kinds of data to be collected, data that shed light on the position of
domestic firms, establishments, and workers in GVCs. As a partial solution to this data gap, we
advocate the collection of establishment-level economic data according to a standardized set of
generic business functions. We share with Lall the desire to move beyond given industry and
product classifications, and to create broad analytical frameworks and data collection tools to
examine aspects of global integration that cut across specific industries and countries. The GVC
framework, the business function scheme, and Lall’s technological classification of exports are
all attempts to create intellectual tools and data classification schemes of exactly this sort.
59
References
Amsden, A.H. 1989. Asia’s Next Giant: South Korea and Late Industrialization. New York:
Oxford University Press.
Arndt, S., and H. Kierzkowski, eds. 2001. Fragmentation: New Production Patterns in the World
Economy. Oxford: Oxford University Press.
Bair, J., and G. Gereffi. 2001. “Local Clusters in Global Chains: The Causes and Consequences
of Export Dynamism in Torreon’s Blue Jeans Industry.” World Development 29(11):
1885–1903.
Barham, N., S. Dörry, and E.W. Schamp. 2007. “Relational Governance and Regional
Upgrading in Global Value Chains—The case of Package Tourism in Jordan.” Die Erde
138(2): 169–186.
Berger, S., and R. Dore, eds. 1996. National Diversity and Global Capitalism. Ithaca, NY:
Cornell University Press.
Berger, S., and the MIT Industrial Performance Center. 2005. How We Compete. New York:
Doubleday.
Bernard, A.B., J.B. Jensen, and P.K. Schott. 2005. “Importers, Exporters and Multinationals: A
portrait of firms in the U.S. that trade goods.” NBER Working Paper No. 11404.
Cambridge, MA: National Bureau of Economic Research.
Borrus, M., D. Ernst, and S. Haggard, eds. 2000. International Production Networks in Asia.
London and New York: Routledge.
Bradach, J., and R. Eccles. 1989. “Price, Authority, and Trust: From Ideal Types to Plural
Forms.” Annual Review of Sociology 15: 97–118.
Brown, S. 2008. “Business Processes and Business Functions: A New Way of Looking at
Employment.” Monthly Labor Review 131(12): 51–70.
Brülhart, M. 2008. “An Account of Inter-industry Trade, 1962–2006.” Research paper series in
Globalisation, Productivity, and Technology, 2008/08. Nottingham: Leverhulme Centre
for Research on Globalisation and Economic Policy, University of Nottingham.
Dean, J.M., K.C. Fung, and Z. Wang. 2007. “Measuring the Vertical Specialization in Chinese
Trade.” Working Paper No. 2007-01-A. Washington, DC: U.S. International Trade
Commission, Office of Economics
Deyo, F.C., ed. 1987. The Political Economy of the New Asian Industrialism. Ithaca and London:
Cornell University Press.
61
Doh, J. 2005. “Offshore Outsourcing: Implications for International Business and Strategic
Management Theory and Practice.” Journal of Management Studies 42(3): 695–704.
Dolan, C., and J. Humphrey. 2000. “Governance and Trade in Fresh Vegetables: The Impact of
UK Supermarkets on the African Horticulture Industry.” Journal of Development Studies
37(2): 147–176.
Evans, P. 1995. Embedded Autonomy: States and Industrial Transformation. Princeton, NJ:
Princeton University Press.
Evgeniev, E. 2008. Industrial and Firm Upgrading in the European Periphery: The Textile and
Clothing Industry in Turkey and Bulgaria. Bulgaria: Professor Marin Drinov Academic
Publishing House.
Feenstra, R. 1998. “Integration of Trade and Disintegration of Production in the Global
Economy.” Journal of Economic Perspectives 12(4): 31–50.
Feenstra, R., and G. Hansen. 2004. “Intermediaries in Entrepôt Trade: Hong Kong Re-exports of
Chinese Goods.” Journal of Economics and Management Strategy 13(1): 3–35.
———. 2005. “Ownership and Control in Outsourcing to China: Estimating the Property-Rights
Theory of the Firm.” Quarterly Journal of Economics 120: 729–761.
Feenstra, R.C., and G.H. Hamilton. 2006. Emergent Economies, Divergent Paths: Economic
Organization and International Trade in South Korea and Taiwan. Cambridge, UK:
Cambridge University Press.
Feenstra, R.C., and B.J. Spencer. 2005. “Contractual versus Generic Outsourcing: The Role of
Proximity.” NBER Working Paper No. 11885. Cambridge, MA: National Bureau of
Economic Research.
Finger, J. 1975. “Trade Overlap and Intra-Industry Trade.” Economic Inquiry 13(4): pp. 581–89.
Fröbel, F., J. Heinrichs, and O. Kreye. 1980. The New International Division of Labor,
Cambridge: Cambridge University Press.
Galvin, P., and A. Morkel (2001). “The Effect of Product Modularity on Industry Structure: The
Case of the World Bicycle Industry.” Industry and Innovation 8(1): 31–47.
Gereffi, G. 1994. “The Organization of Buyer-Driven Global Commodity Chains: How U.S.
Retailers Shape Overseas Production Networks.” In Commodity Chains and Global
Capitalism, G. Gereffi and M. Korzeniewicz, eds. Westport, CT: Praeger, pp. 95–122
———. 1996. “Commodity Chains and Regional Divisions of Labor in East Asia”, Journal of
Asian Business, 12(1): 75–112.
———. 1999. “International Trade and Industrial Upgrading in the Apparel Commodity Chain.”
Journal of International Economics 48(1): 37–70.
62
———. 2005. “The Global Economy: Organization, Governance, and Development.” In The
Handbook of Economic Sociology, 2nd ed, N.J. Smelser and R. Swedberg eds. Princeton,
NJ: Princeton University Press and Russell Sage Foundation, pp. 160–182.
Gereffi, G., J. Humphrey, and T. Sturgeon. 2005. “The Governance of Global Value Chains.”
Review of International Political Economy 12(1): 78–104.
Gower, A., and M. Cusumano. 2002. Platform Leadership: How Intel, Microsoft, and Cisco
Drive Industry Innovation. Cambridge, MA: Harvard Business School Press.
Grossman, G.M., and E. Helpman. 1991. “Quality Ladders in the Theory of Growth.” Review of
Economic Studies 58: 53–61.
Hall, P.A., and D. Soskice, eds. 2001. Varieties of Capitalism: The Institutional Foundations of
Comparative Advantage. Oxford: Oxford University Press.
Hamilton, G.H., and G. Gereffi. 2008. “Global Commodity Chains, Market Makers, and the Rise
of Demand-Responsive Economies.” In Frontiers of Commodity Chain Research, J. Bair,
ed. Palo Alto, CA: Stanford University Press.
Harrison, A.E., and M.S. McMillan. 2006. “Outsourcing Jobs? Multinationals and U.S.
Employment.” NBER Working Paper No. 12372. Cambridge, MA: National Bureau of
Economic Research.
Hausmann, R., J. Hwang, and D. Rodrik. 2006. “What You Export Matters.” Working Paper No.
123. Cambridge, MA: Center for International Development, Harvard University,
December 2005 (revised March 2006).
Head, K., J. Ries, and B. Spencer. 2004. “Vertical Networks and U.S. Auto Parts Exports: Is
Japan Different?” Journal of Economics and Management Strategy 13(1): 37–67.
Hummels, D., J. Ishii, and K.-M. Yi. 2001. “The Nature and Growth of Vertical Specialization in
World Trade.” Journal of International Economics 54(1): 75–96.
Humphrey, J. 2003. “Globalization and Supply Chain Networks: The Auto Industry in Brazil and
India.” Global Network 3(2): 121–141.
Huws, U., and S. Dahlmann. 2004. Outsourcing of ICT and Related Services in the EU: A Status
Report. European Foundation for the Improvement of Living and Working Conditions.
Luxembourg, Office for Official Publications of the European Communities.
www.eurofound.europa.eu/emcc/publications/2004/ef04137en.pdf (accessed June 1, 2010).
Jarmin, R.S., and J. Miranda. 2002. “The Longitudinal Business Database.” U.S. Census Bureau
Working Paper No. 02-17. Washington, DC: Census Bureau.
http://ideas.repec.org/p/cen/wpaper/02-17.html (accessed June 1, 2010).
Kawakami, M., and T. Sturgeon, eds. Forthcoming. The Dynamics of Local Learning in Global
Value Chains: Experiences from East Asia. London: Palgrave.
63
Kimura, S. 2007. The Challenges of Late Industrialization: The Global Economy and the
Japanese Commercial Aircraft Industry. Basingstoke and New York: Palgrave Macmillan.
Kletzer, L. 2002. Imports, Exports, and Jobs: What Does Trade Mean for Employment and Job
Loss? Kalamazoo, MI: W.E. Upjohn Institute for Employment Research.
Krugman, P. 1991. “Increasing Returns and Economic Geography.” Journal of Political
Economy 99: 483–499.
Lall, S. 2000. “The Technological Structure and Performance of Developing Country
Manufactured Exports, 1985–98.” Oxford Development Studies 28(3): 337–369.
Lall, S., and M. Albaladejo. 2004. “China’s Competitive Performance: A Threat to East Asian
Manufactured Exports?” World Development 32(9): 1441–1466.
Lall, S., J. Weiss, and J. Zhang. 2005. “The ‘Sophistication’ of Exports: A New Measure of
Product Characteristics.” Oxford, UK: Queen Elizabeth House Working Paper Series,
123. January.
Lee, J.R., and J.S. Chen. 2000. “Dynamic Synergy Creation with Multiple Business Activities:
Toward a Competence-Based Business Model for Contract Manufacturers.” In Advances
in Applied Business Strategy, 6A (Theory Development for Competence-based
Management), Ron Sanchez and Aime Heene, eds. Stamford, CT: JAI Press.
Lewin, A.Y., S. Massini, and C. Peeters. 2009. “Why Are Companies Offshoring Innovation?
The Emerging Global Race for Talent.” Journal of International Business Studies 40(6):
pp. 901–925.
Linden, G., K.L. Kraemer, and J. Dedrick. 2007. “Who Captures Value in a Global Innovation
System? The Case of Apple’s iPod.” Personal Computing Industry Center Working
Paper. Irvine, CA: University of California at Irvine.
http://pcic.merage.uci.edu/papers.asp (accessed June 1, 2010).
Lloyd, P.J., and H-H. Lee, eds. 2002. Frontiers of Research in Intra-Industry Trade, London:
Palgrave Macmillan.
Maule, C. 2006. “Integrative Trade: Issues for Trade Analysis, Statistics and Policy.” Paper
prepared for the “CTPL Conference on Integrative Trade between Canada and the United
States—Policy Implications,” held in on December 6.
Neilson, J., and B. Pritchard. 2009. Value Chain Struggles: Institutions and Governance in the
Plantation Districts of South India. West Sussex, UK: Wiley-Blackwell.
Ng, F., and A. Yeats. 1999. “Production Sharing in East Asia. Who Does What for Whom, and
Why?” World Bank Policy Research Working Paper No. 2197. Washington, DC: World
Bank.
64
Nielsen, P.B., ed. 2008. International Sourcing—Moving Business Functions Abroad, Statistics
Denmark, Copenhagen. www.dst.dk/publ/InterSourcing (accesed June 1, 2010).
Pietrobelli, C., and R. Rabellotti. 2007. Upgrading to Compete: Global Value Chains, Clusters
and SMEs in Latin America. Cambridge, MA: Harvard University Press.
Pine, J.P., and S. Davis. 1999. Mass Customization: The New Frontier in Business Competition.
Cambridge, MA: Harvard Business School Press.
Piore, M., and C. Sabel. 1984. The Second Industrial Divide. New York: Basic Books.
Porter, M. 1985. Competitive Advantage. New York: Free Press.
Rasiah, R. 2009. “Sanjaya Lall: The Scholar and the Policy Advisor.” International Journal of
Institutions and Economies 1(1): 8–22.
Saxenian, A. 2006. The New Argonauts: Regional Advantage in a Global Economy. Cambridge,
MA: Harvard University Press.
Schmitz, H. 1999. “Global Competition and Local Cooperation: Success and Failure in the Sinos
Valley, Brazil.” World Development 27(9): 1627–1650.
Sturgeon, T. 2002. “Modular Production Networks: A New American Model of Industrial
Organization.” Industrial and Corporate Change 11(3): 451–496.
———. 2007. “How Globalization Drives Institutional Diversity: The Japanese Electronics
Industry’s Response to Value Chain Modularity.” Journal of East Asian Studies 7(1): 1–
34.
Sturgeon, T., and J.-R. Lee. 2005. “Industry Co-evolution: A Comparison of Taiwan and North
American Electronics Contract Manufacturers.” In Global Taiwan: Building Competitive
Strengths in a New International Economy, Suzanne Berger and Richard K. Lester, eds.
Armonk, NY: M.E. Sharpe.
Sturgeon, T., and O. Memedovic. Forthcoming. “Measuring Global Value Chains: Intermediate
Goods Trade, Structural Change and Compressed Development.” UNIDO Working
Paper. Vienna: United Nations Industrial Development Organization.
Sturgeon, T., and R. Florida. 2004. “Globalization, Deverticalization, and Employment in the
Motor Vehicle Industry.” In Locating Global Advantage: Industry Dynamics in a
Globalizing Economy. M. Kenny, with R. Florida, eds. Palo Alto, CA: Stanford
University Press.
Sturgeon, T., J. Van Biesebroeck, and G. Gereffi. 2008. “Value Chains, Networks and Clusters:
Reframing the Global Automotive Industry.” Journal of Economic Geography 8(3): 297–
321.
65
Swenson, D. 2005. “Overseas Assembly and Country Sourcing Choices.” Journal of
International Economics 66(1): 107–130.
Vind, I., and N. Fold. 2007. “Multi-level Modularity vs. Hierarchy: Global Production Networks
in Singapore’s Electronics Industry.” Danish Journal of Geography, 107(1): 69–83.
Wade, R. 1990. Governing the Market: Economic Theory and the Role of Government in East
Asian Industrialization. Princeton, NJ: Princeton University Press.
Whittaker, D.H, T. Zhu, T. Sturgeon, M.H. Tsai, and T. Okita. Forthcoming. “Compressed
development.” Studies in Comparitive International Development.
World Bank. 1993. The East Asian Miracle: Economic Growth and Public Policy. New York:
World Development Indicators. 2007. New York: World Bank. http://web.worldbank.org. Cited
in Federation of International Trade Associations,
http://fita.org/countries/malaysia.html?ma_rubrique=economie.
Yeats, A. 2001. “Just How Big Is Global Production Sharing?” In Fragmentation: New
Production Patterns in the World Economy, S. Arndt and H. Kierzkowski, eds. Oxford:
66
SESSION 2: MEASURING THE IMPACT OF
OFFSHORING ON THE LABOR MARKET
CHAIR: Janet Norwood* (NAPA)
Addressing the Demand for Time Series and Longitudinal Data on Occupational Employment,
Katharine Abraham (University of Maryland) and
James R. Spletzer (Bureau of Labor Statistics)
Measurement Issues, Lori Kletzer (University of California-Santa Cruz and
Peterson Institute for International Economics)
DISCUSSANT: Timothy Sturgeon (MIT)
* National Academy Fellow
67
Addressing the Demand for Time Series and Longitudinal
Data on Occupational Employment
*
Katharine G. Abraham
University of Maryland and NBER
James R. Spletzer
U.S. Bureau of Labor Statistics
October 22, 2009
Revised February 26, 2010
*
Without implicating them in any way, we are grateful to Laurie Salmon, Dixie Sommers, and George Stamas,
Office of Employment and Unemployment Statistics, Bureau of Labor Statistics, for useful conversations and
background information concerning the Occupational Employment Statistics survey.
69
Among the important potential effects of increased offshoring are changes in the occupational
composition of U.S. employment. To the extent that firms choose to shift particular tasks to
workers located overseas, domestic employment in the occupations that perform those tasks can
be expected to decline or to grow less rapidly than would otherwise have been the case. It has
been suggested that time-series data on occupational employment by industry could be useful for
studying these effects in the aggregate. Similarly, longitudinal data on the mix of jobs at
individual enterprises could be useful for better understanding the dynamics of outsourcing and
offshoring at the level of the individual firm. The development of this sort of information has
been recommended as part of a broader set of needed improvements in the data available for the
study of off-shoring (see, for example, Sturgeon [2006] and National Academy for Public
Administration [2006]).
Concerns about sample size and the accuracy with which household respondents report their
occupations and industries, together with the fact that household survey data generally cannot be
used to study the evolution of employment at individual firms, have led analysts interested in the
effects of offshoring on domestic employment to focus on employer-provided employment data.
The Occupational Employment Statistics (OES) survey conducted by the Bureau of Labor
Statistics (BLS) is a large employer survey that, each year since 1988, has collected detailed
information on employment by occupation.
The OES is designed to produce detailed point-in-time estimates of staffing patterns and wages,
not to produce occupational employment time series or to support the analysis of changes in the
occupational composition of employment at individual workplaces. For some applications, such
as the use of OES employment data to determine weights in the BLS National Compensation
Survey program or the use of OES wage data by the Employment and Training Administration to
administer the H-1B visa program, having data that could be compared over time would not be
especially valuable. Many users of OES survey data, however, clearly would benefit from data
designed to support cross-year comparisons. For example, annual data designed to track
trajectories in staffing patterns would be of great value to the BLS Occupational Employment
Projections (OEP) program. Similarly, access to information on trends in occupational staffing
patterns and wages would help those who use the data for workforce development, career
counseling, and career planning purposes. Researchers studying organizational behavior, the
sources of productivity growth, and other topics could benefit from data that allowed them to
track staffing patterns at the level of the individual establishment or individual enterprise.
The purpose of this paper is to evaluate the OES survey as a source of time-series and/or
longitudinal data on employment by occupation. This sort of information would be useful not
only for identifying offshoring activity and study its impacts—topics that are the focus of the
present conference—but for a variety of other purposes as well. In the next section we briefly
describe the OES survey. We then discuss the feasibility of using the historical OES data to
construct occupational employment time series or for longitudinal analysis at either the
establishment or the enterprise level. Unfortunately, the existing survey design and the
management approach dictated by current program objectives make the data poorly suited for the
analysis of trends, especially over short time intervals. The fourth section considers how the
OES survey might be reconfigured to produce reliable annual time-series data and support
analysis at the individual establishment or enterprise level. A necessary step would be to collect
70
data from some subset of establishments every year, rather than only once every three years as is
the current practice, but other changes in the survey also would be required. Some concluding
thoughts and observations are offered in the final section.
THE OES SURVEY
The OES survey is an ongoing mail survey conducted by the BLS in collaboration with its state
partners. It covers all industries exclusive of agriculture. Prior to 1996, industries were surveyed
on a three-year rotating cycle. Since 1996, each year’s sample has covered establishments across
all industries, but except for an annual enumeration of federal and state government
establishments, it is still the case that even very large establishments are surveyed only once
every three years.
The OES sample is designed to support cross-sectional estimates of staffing patterns developed
from data collected over a three-year period. Through 2001, estimates were based on three
annual panels, each consisting of approximately 400,000 establishments; within each panel,
establishments were assigned an October, November, or December reference date. In 2002, the
survey transitioned to a design in which estimates are based on six semiannual panels, with each
panel consisting of approximately 200,000 establishments assigned either a May or a November
reference date. The May 2008 published estimates, for example, rest on data collected for
November 2005, May 2006, November 2006, May 2007, November 2007, and May 2008.
Estimates are benchmarked to the average of the most recent May and November employment
levels.
Since 1996, the OES has collected information on occupational wages in addition to
occupational employment. Establishments selected for the OES are asked to report employment
in each cell of a matrix in which the rows refer to different occupations and the columns to wage
intervals. Generally, for firms with 20 or more employees, the survey forms contain between 50
and 225 occupations, depending on the industry of the establishment completing the form. Prior
to 2000, employers receiving these forms were asked to list numerically significant or new
occupations that could not be reported in a detailed occupation and therefore were reported in an
“all other” residual category. This information was used in revising the survey forms for later
years. Beginning in 2000, employers have been asked to provide detailed occupational
information for workers who cannot be placed in one of the listed occupations.
Since 1999, small employers have received a shorter unstructured form that contains no list of
likely occupation titles; rather, the employer is asked to provide a brief description of each
occupation represented in the establishment’s workforce. The information on these forms is
coded into occupational categories by survey staff in the state agencies.22 Multiestablishment
firms may request that their data be collected through the firm’s corporate headquarters rather
than directly from individual establishments. This is referred to as central office collection
(COC). COC reporters provide the OES program with electronic records containing job title and
22
Prior to 1999, several states developed their own unstructured short forms that were used to collect data from some
small employers, but this was not a part of the formal survey protocol. Beginning in 2004, states were given the
discretion to send unstructured forms to establishments with up to 49 employees.
71
wage information for their employees. The OES staff then builds crosswalks for coding these
firms’ data into Standard Occupational Classification (SOC) occupations and OES wage
intervals.
Approximately 80 percent of establishments sampled for the OES survey provide usable
responses; on an employment-weighted basis, the survey response rate is approximately 75
percent. Nearest neighbor hot-deck procedures, which take data from another similar
establishment, are used to impute missing employment information for establishments that do
not respond. Missing wage distributions also are imputed using distributions for similar
establishments.
OES survey data are published by occupation, industry, and area. Employment and wage
estimates are produced for as many as about 800 occupations, both nationally and for states,
metropolitan areas and other geographic areas. In addition, national occupational employment
and wage estimates are available for specific industries. BLS does not publish occupation by
industry data below the national level, and suppression of sparsely populated cells is common in
both the national by-industry tables and the cross-industry tables for subnational areas. The OES
program switched from its own survey-specific occupation coding system to the SOC system in
1999 and from the Standard Industrial Classification (SIC) system to the North American
Industry Classification System (NAICS) in 2002.
CAN EXISTING OES DATA BE USED TO CONSTRUCT ANNUAL TIME SERIES?
Several features of the current OES design make it difficult to use the existing data to construct
single-year estimates of employment by detailed occupation, even at the national level. The
current design calls for estimates to be produced using three years of data, and the existing
survey weights are not suitable for the production of annual estimates. It is possible to construct
annual estimation weights, but because (except in federal government) even very large units are
surveyed only once every three years, annual estimates tend to be quite variable, especially for
smaller employment cells. The significant breaks in both occupation and industry classification
caused by the adoption of the SOC and the NAICS are another problem. Finally, other changes
in survey operations associated with the adoption of the SOC have affected the comparability of
the OES estimates over time. To preview our conclusions, we believe that it is possible to use
existing OES survey data to construct national occupational time series that are suitable for some
analytic purposes, but that these estimates have serious limitations for studying offshoring and its
effects.
Lack of Weights for Annual Estimates
The weights used to produce official OES estimates are constructed at the level of cells defined
on the basis of industry, establishment size, and geography. As noted above, the sample units
used to produce each set of estimates are divided into panels spread across three years of data
collection. Each sampled establishment is assigned a current weight that reflects its probability
72
of selection into a particular panel.23 If every cell in a panel contained at least one establishment,
the weighted sum of employment calculated for an industry using the current weights would be
approximately equal to total national employment in the industry as of the panel reference
date(s). There are, however, a very large number of OES sampling cells—as of 2004, the survey
was stratified by 343 industries, seven establishment size classes, and 686 metropolitan or
balance-of-state geographic areas—and individual panels contain a significant number of empty
cells. Because employment in the cells that happen to be empty is not represented, using the
current weights to estimate employment in an industry based on the responses to any single panel
yields an estimate that lies significantly below the industry’s true employment level.24
Working with OES data for the private sector over the period 1996–2004, Abraham and
Spletzer (forthcoming) developed weight adjustment factors to be applied to the OES current
weights that are calculated as follows:
(1)
ADJFACTOR1 jt =
E CES
jt
∑ CURRWT
OES
ijt
EijtOES
,
i
where ADJFACTOR1 is the industry weight adjustment factor, E is employment from either the
monthly Current Employment Statistics (CES) survey or the OES survey, CURRWT is the
current weight from the OES data file, i indexes individual establishments, j indexes detailed
industries, and t indexes years. Estimates produced from the OES microdata using weights equal
to the product of ADJFACTOR1, and CURRWT reproduce CES national industry employment
trends.
In all years, industry weight adjustment factors were calculated at the most detailed industry
level for which sample data were available. The SIC classification structure in use through 2001
included 934 detailed industries. In 1996, taking that year as an example, weight adjustment
factors were calculated at the four-digit (most detailed) level for 310 industries, representing 34.0
percent of employment; at the three-digit level for 383 industries, representing 37.6 percent of
employment; at the two-digit level for 225 industries, representing 27.8 percent of employment;
and at the one-digit level for 16 industries, representing 0.6 percent of employment. The NAICS
structure adopted in 2002 includes 1,171 detailed industries. In 2004, weight adjustment factors
were calculated at the five-digit (most detailed) level for 424 industries, representing 36.1
percent of employment; at the four-digit level for 520 industries, representing 47.1 percent of
employment; at the three-digit level for 172 industries, representing 9.8 percent of employment;
and at the two-digit level for 55 industries, representing 7.0 percent of employment.
A further weighting concern is that, although the true distribution of employment by size of
establishment appears to have been very stable from 1996 through 2004, the distributions in the
data collected for the OES vary considerably from year to year. Factors that appear to have
contributed to this variability include the uneven distribution of the largest (certainty) units
across panels; the effects of a 1999 experiment carried out in selected states to determine the
23
The current weights also incorporate adjustments for differences between the way a unit was sampled and the way
it was reported (e.g., one establishment at a company sampled but data reported for several establishments together).
24
In the official estimates, which are based on three years of data, this is not generally a problem because data at the
detailed cell level are reweighted to account for the number of panels in which each cell is represented.
73
feasibility of collecting data from all certainty establishments every year; and the introduction of
establishments with 1–4 employees into the survey sample in 1998 (these very small
establishments previously had been represented by establishments with 5–9 employees). A
second weight adjustment factor was developed to calibrate the share of employment in broad
industries accounted for by each of nine establishment size classes to the average share for that
size class across the OES benchmark data files for 1998, 2001, and 2004.25
(2)
ADJFACTOR 2kst =
AVESHAREksBMK
,
OES
SHAREkst
where ADJFACTOR2 is the size class weight adjustment factor, AVESHARE is the average share
of employment accounted for by the designated size class in the benchmark data, SHARE is the
current year share in the OES data, k indexes broad industry, s indexes establishment size class,
and t indexes year. Applying both the industry and the size class adjustment factors yields
(3)
OES
FINALWTijkst = ADJFACTOR1 jt x ADJFACTOR 2kst x CURRWTijkst
.
Anyone interested in using the historical OES data to construct an annual time series would need
to apply some similar procedure to produce weights suitable for annual estimates.
Variability in Annual Estimates
In cells defined at the national level using broad industries and occupations, employment
estimates calculated using the adjusted weights just described seem generally to behave very
sensibly (see Appendices C and D in Abraham and Spletzer forthcoming). Because even very
large units in the private sector are surveyed only once every three years, however, annual
estimates of employment for detailed occupations, estimates for subnational areas and/or
estimates for occupation by industry are likely to behave more erratically. To illustrate the
potential instability in estimates for small domains, we used the adjusted weights just described
to construct annual time series for the 10 occupations identified by Jensen and Kletzer
(forthcoming) as most offshorable. These are shown in Figure 1. Each of the 10 panels in the
figure contains two employment series—one created from year-specific OES microdata with
weights adjusted using the method just described, and one created from the November OES press
releases posted to the BLS Web site.26 We should note that using the published OES estimates in
this way is not recommended by the BLS, which states on its Web site that it “does not use or
encourage the use of OES data for time series analysis”
(http://www.bls.gov/oes/oes_ques.htm#Ques29). Still, we believe that the comparisons shown in
the figure are informative.
25
The OES survey data are benchmarked to the quarterly Census of Employment and Wages. Our analysis rests on
OES benchmark data files for the years 1998, 2001, and 2004. The size class distributions observed across these
three years are very similar.
26
Through 2001, all OES data were collected with a reference date in October, November or December. In 2002, the
OES switched to May and November reference dates. For these later years, in an effort to avoid problems of
comparability associated with seasonal differences in staffing patterns at different times of the year, we used only
the data from the November panel. The 2003 and 2004 press releases report statistics benchmarked to average
employment by industry for the most recent May and November and there may be an issue of comparability
between the published estimates for these years as compared to earlier years.
74
The estimates in Figure 1 cover the years from 1999, the year that the SOC was adopted in the
OES program, through 2004. One difference between the two series is that the microdata
estimates cover only employment in the private sector, while the published estimates cover
employment in federal, state, and local government, as well as the private sector. While this
difference in coverage has a noticeable effect on the level of some of the series—most obviously,
the series for statisticians and medical transcriptionists—it should not much affect their year-toyear variability. More importantly for the series’ variability, the microdata estimates are based
on establishments that, from 1999 through 2001, represented only about one-third of the full
private sector OES sample and, from 2002 through 2004, only about one-sixth of the full private
sector sample. The time series created with annual microdata are considerably more volatile
than the time series created from the published data. Several of the occupational time series
based on annual microdata—those for mathematical technicians; credit authorizers, checkers and
clerks; biochemists and biophysicists; title examiners; weighers; and actuaries—show sharp
changes from one year to the next that are not apparent in the published estimates based on
multiple years of data. For 9 of the 10 occupations in Figure 1—excluding only statisticians—
the variance of the series created from the annual microdata is higher than that for the published
estimates. Figure 1 suggests that employment time series for detailed occupations that are
created from single-year OES microdata are likely to be highly volatile, making them
problematic for policy analysis of the effects of offshoring. Increases in the size of the OES
sample would be needed to reduce the variance of annual employment estimates.
Breaks in Occupation and Industry Classification Systems
Breaks in both the occupation and industry classification structures are an additional barrier to
using the historical OES data to produce detailed annual time series. As already noted, prior to
1999, the OES used its own classification structure; the Standard Occupational Classification
(SOC) was introduced in 1999. The NAICS replaced the older SIC system in 2002. In both
cases, the new classification structure was very different from the old. Of the 769 detailed
occupations included in the SOC when it was introduced in 1999, only 374 could be crosswalked directly to occupations that previously existed in the old OES classification structure
(BLS 2001a, pp. 24 and 175). During the transition to NAICS at the BLS, only about half of
establishments could be assigned NAICS codes based on their SIC classification (Mikkelson,
Morisi, and Stamas 2000).
Comparisons at an aggregated level seem more feasible across the classification structure breaks
than do more disaggregated comparisons. Matthew Dey of the BLS, for example, has developed
a concordance that links cells defined using 19 aggregated occupations and 13 aggregated
industries that appear to be reasonably consistent across the breaks in classification system, and
Abraham and Spletzer (forthcoming) use a modified version of the Dey concordance. There are
detailed occupations within these larger groupings that are defined in the same way in the SOC
as in the older OES classification structure, but as already noted there are also many detailed
occupations for which no direct linkage is possible. The OES data were never dual coded using
the SOC and OES occupational classification structures, so in cases where occupations are not
comparable it is difficult to relate the new SOC occupations to the older OES occupations.
The best option for extending the OES annual estimates for detailed occupations back through
time likely would be to plot the employment series for each of the occupations and retain those
75
occupations for which there is no evidence of a discontinuity between 1998 and 1999 that might
indicate a lack of comparability in how the occupation was defined. Where no direct match at
the detailed occupation level exists, occupations formed by combining detailed occupations in an
appropriate fashion could be evaluated similarly. Drawing clear conclusions could be difficult,
however, because of the substantial underlying variability in annual estimates based on the OES
microdata that has already been discussed. Moreover, since industries were surveyed on a threeyear rotating cycle prior to 1996, even under the best-case scenario, it would be possible to
extend true annual series by only three years, back to 1996.
Other Comparability Issues
In addition to a new coding structure, the SOC also introduced a set of principles intended to
guide the classification of workers (OMB 2000). An important principle is that only individuals
who devote at least 80 percent of their time to management activities are to be classified as
managers. To implement this SOC guidance, the OES introduced new edit checks to flag
establishments that reported employment in a management occupation (e.g., financial manager)
without reporting employment in any of the expected subordinate occupations (e.g., financial
specialists or clerks). A second set of edit checks was developed to flag establishments with an
excessive number of managers. Both sets of edit checks were applied in a limited fashion in
1999 and phased in more fully over the following years.
Implementation of the SOC also included new training designed to explain its structure and
coding principles to program staff. Staff who attended SOC training courses in 1999 and
subsequent years were instructed that management jobs reported on establishment schedules that
did not include an intervening layer of supervision generally would need to be recoded as
something else. Management jobs recoded in accord with this advice typically were shifted
either to one of the professional occupations or to a first-level supervisor occupation. Because
survey program staff code all of the occupations reported on unstructured survey forms
submitted by small establishments, the introduction of the unstructured forms as an option in
1999 may have amplified the effects of the SOC training on the OES management employment
series. The “rule” that no job should be coded as a management position unless the schedule also
includes a first-level supervisor position is easy to apply and seems to have been embraced as a
guide to coding the unstructured schedules.
Our best assessment is that the combined effect of these changes was to reduce management
employment by very roughly 2 million jobs between 1998 and 2001, with these jobs then
assigned instead to other job categories. Although perhaps less important for the analysis of
offshoring than for some other purposes, this is nonetheless an additional barrier to comparing
OES estimates over time. In our earlier work (Abraham and Spletzer forthcoming), we
developed a procedure for “reverse-engineering” the new coding rules that involved reclassifying
a sufficient number of nonmanagement jobs in 1999 and later years as management positions to
offset the sharp decline in management employment that is evident in the unadjusted data
between 1999 and 2001. This effort was unavoidably crude. Further, going forward, insofar as
76
it involves putting new OES data onto the old basis rather than putting old data on the new basis,
the application of this procedure would inevitably become less and less appealing.27
New SOC training introduced in 2007 makes clear that someone might legitimately be
performing management duties without there being an intervening layer of supervision between
them and their subordinates. While this change was made to improve coding accuracy, going
forward, it too may adversely affect the comparability of the OES data over time.
Problems with Coverage of Units Surveyed
A further limitation of the historical OES data is that the survey sample is not designed to
support longitudinal analysis. As has been noted, establishments are asked to report on a threeyear cycle. Large establishments responding to the survey in year t are likely also to have
responded in year t-3. Because even large establishments are observed only once every three
years, there may be a long lag before important changes in staffing patterns are captured.
Further, the establishment is not the obvious unit of analysis for identifying and tracing the
effects of offshoring. In a large corporation, sourcing decisions are likely to be made at the
corporate level rather than the establishment level, and a decision to offshore work could take the
form of shuttering an entire establishment, rather than transferring portions of the work
performed at an individual establishment to another company. In this case, offshoring could not
be identified through an establishment-level analysis, but only through an examination of
changes in staffing patterns for the company as a whole.
To the extent that the appropriate level of analysis is the company rather than the establishment,
the OES survey suffers from the further limitation that data generally are not collected for all of
the establishments at a firm (National Academy of Public Administration 2006). Over the course
of a three-year period, all establishments large enough to belong to the survey’s certainty strata
are asked to complete an OES questionnaire, but large firms include many small establishments,
and only a fraction of these small establishments would be surveyed even once in a three-year
cycle. Among firms in the United States with more than 10,000 employees, for example, in May
2006 there were 377,484 establishments with fewer than 250 employees, accounting for 53.6
percent of these firms’ employment; 7,369 establishments with 250–499 employees, accounting
for an additional 13.9 percent of the firms’ employment; and 2,727 establishments with 500–999
employees, representing a further 10.1 percent of the very large firms’ employment.28
27
The specific jobs that formerly might have been categorized as management jobs were identified by looking for
the highest-paid nonmanagement jobs in establishments with too few managers, based on historical patterns. This
procedure took jobs from many different occupations and reclassified them as management positions (many to one).
We were not able to devise a methodology for putting the old data on the new basis. That would have involved
reassigning management positions to many different possible alternate occupations (one to many), and we did not
believe we had a sound basis for making such assignments.
28
These figures and others cited in the text are based on tabulations of the Business Employment Dynamics
database, which is based on the Quarterly Census of Employment and Wages (QCEW) data file and are designed
only to be illustrative.
77
REDESIGNING THE OES FOR TIME SERIES AND LONGITUDINAL ANALYSIS
To produce reliable annual time series, the OES sample would need to be augmented with units
that are surveyed every year rather than only once every three years. Being able to track changes
in staffing patterns at the establishment level might be of some interest, though as just discussed,
important developments could be missed with an establishment-level focus. To the extent that
analysts are interested in using enterprise-level data to study the effects of offshoring, it would be
desirable for at least some firms to provide comprehensive data for all of their establishments. In
addition to changes in the design of the survey sample to support time series and/or longitudinal
analysis, other changes in the survey’s focus also would be necessary.
Sample Redesign Options
We consider three possible options for augmenting the OES sample. The first is intended to be
responsive to the demand for a sample that could support annual OES time series; the second
suggests an approach to collecting data suitable for longitudinal analysis; and the third is a
hybrid of the first two approaches. Many variations on these approaches can be imagined, and
we do not mean to suggest that the specific options we describe are the only ones that are
possible. Further, because it is the information we were able to access given our time
constraints, we use data on number of firms and number of establishments from 2006 in our
rough cost calculations, though since the size distribution of establishments is very stable over
time and total employment has changed little on net between 2006 and the present, we do not
believe our answers would have been much different had we been able to use 2009 data for these
calculations. Our intent in any case is to stimulate thinking about sample design alternatives and
the rough magnitude of the costs that might be associated with different choices, rather than to
recommend a specific plan and attach a specific dollar cost figure to that plan. The rough dollar
cost figures we present refer only to the direct costs of additional data collection and do not
include the costs of other staffing necessary to edit and process the additional data collected. All
of the options considered focus on increases in the size of the sample for the private sector, as
that is where we would expect the effects of offshoring to be manifested.
Option 1: Survey all large private sector establishments every year. One way to increase
the stability of annual OES estimates over time would be to survey all establishments with more
than some threshold level of employment every year. Suppose, for example, that the survey
were redesigned to collect data from all establishments with 250 or more employees every year.
As of May 2006, the universe of private sector establishments eligible for inclusion in the OES
included 30,639 establishments with 250–499 employees, 10,894 establishments with 500–999
employees, and 5,470 establishments with 1,000 or more employees. Since even units included
in the current certainty strata are surveyed only once every three years, surveying all units with
250 or more employees every year would represent a significant increase in the annual survey
sample.
Most of the work of collecting OES data is done by the states. Payments to the states for their
work on the OES program totaled $21.5 million in FY2009 (the fiscal year in which the 2008
estimates were published). This figure includes some state overhead expenses and excludes some
modest expenses for data collection work performed by the BLS national office staff, but for
present purposes, we treat the payments made to the states as the cost of OES data collection.
78
The money awarded to states is allocated using a formula that takes into account the number of
establishments surveyed in the state, the size distribution of those establishments, the number of
publication areas in the state, and the average wages of state employees. In this allocation
formula, establishments with 250–499 employees are treated as equivalent to two units,
establishments with 500–999 employees as equivalent to three units and establishments with
1,000 or more employees as equivalent to four units.29 By our calculations, as a very rough
approximation, the current allocation formula implies a cost of $50 per establishment of size up
to 249 employees, $100 per establishment of size 250–499 employees, $150 per establishment of
size 500–999 employees, and $200 per establishment of size 1,000 or more employees.
Applying these cost estimates, to survey all private sector establishments with 250 or more
employees every year would have added roughly $3.9 million dollars to the cost of data
collection for the OES program.30
Without a more detailed analysis that would be beyond the scope of the present exercise, we
cannot say precisely how collecting data from all private sector establishments with 250 or more
employees every year would affect the variance of annual estimates, but the number of
employees for whom data were collected each year clearly would increase significantly. The
current sample includes establishments with total employment of approximately 20.0 million per
year over the three years of the survey cycle; the sample augmentation just described would add
establishments with employment of approximately 19.6 million each year, close to doubling the
employment covered.
Option 2: Survey all private sector establishments in large firms every year. Adoption of
the preceding option for redesign of the OES sample would reduce the variance of annual OES
estimates. Under this plan, however, only large establishments could be followed over time and,
in most cases, data collected in a particular year still would cover only a portion of a firm’s
establishments, making it difficult to use the data for firm-level analysis. If longitudinal analysis
at the firm level is a priority, surveying all of the establishments in firms with more than a
threshold level of employment every year might be an appealing strategy. To the extent that
large firms were willing to provide electronic data files containing information on the job
classifications of all of their employees, collecting data in this way also could yield significant
economies of scale. Indeed, without such economies, this data collection strategy would be
prohibitively expensive.
Turning again to the May 2006 universe listing, using EIN as the firm identifier, we found
322,525 establishments that belonged to the 8,295 firms with 1,000–4,999 employees, 148,211
establishments that belonged to the 964 firms with 5,000–9,999 employees, and 429,140
establishments that belonged to the 677 firms with 10,000 or more employees. Using the perestablishment cost figures cited above, even restricting attention to establishments belonging to
firms with 10,000 or more employees, the costs of data collection would be projected to rise by
29
This information was provided in a personal communication from OES program staff member Laurie Salmon on
September 9, 2009.
30
We should emphasize that the program does not receive a budget based on the number of units included in the
survey sample. Rather, the program receives a total dollar amount of funding that then must be allocated to cover
the various costs of program operation. Similar to other surveys, however, the amount of money needed to maintain
program operations can be expected to grow over time because of growth in wages, salaries and other expenses and
growth in the number of certainty units.
79
$18.7 million dollars per year, an amount that comes close to equaling the program’s entire
current data collection budget.
It seems likely, however, that economies of scale could be realized if data were collected for
entire companies rather than separately for each establishment in the sample. So-called central
office collection (COC) is already in place for some companies who prefer to submit their data in
this way. Estimates of the staffing required to complete the current COC workload compiled by
BLS national and regional office staff (NCS-OES Data Collection and Processing Cost Team
2008) suggest that a reasonable estimate might be that it costs approximately $6,000 per firm to
process these submissions. If COC collection were mandated for all large firms, the per-firm
cost likely would be higher. One reason is that firms for which this is not necessarily the
preferred reporting method would have to be convinced to report in this way. The need to
process all of the firms’ establishments, rather than only selected establishments as at present,
also would raise per-firm costs. To the extent that firms use common job classification systems
across all of their establishments, the effort to assign SOC codes to job titles might not vary a
great deal with the number of included establishments, but there could be other complications.
For example, past experience suggests that there can be problems with matching establishments
listed in the firm records to establishments on the BLS business list and with assigning
appropriate industry and geographic identifiers.
For the purpose of producing a rough data collection budget estimate, suppose that it would cost
$8,000 per COC firm to process electronically submitted data. If the COC universe were
restricted to firms with 10,000 or more employees, this per-firm cost figure implies that total data
collection costs would rise by $3.5 million per year, rather than the $18.7 million implied by the
establishment-by-establishment collection cost model. If we instead assumed a higher figure of
$10,000 per firm, the projected budget increment for added data collection work would be $4.8
million.
Compared to the previous sample redesign option, this option likely would do less to reduce the
variance of annual OES estimates. As of May 2006, 14.0 million people were employed by
establishments that would have been added under this strategy, compared to the 19.6 million
employed by the establishments added under the previous strategy. In addition, to the extent that
establishments of a given firm tend to be similar with respect to their staffing patterns, their
addition to the sample will do less to improve the precision of the aggregate estimates than
would the addition of a similar number of unaffiliated establishments. A major advantage of this
approach is that, because information for all of the establishments at the identified firms would
be collected, the data would be well suited for studying occupational employment trends at the
firm level for the covered firms.
Option 3: Survey all private sector establishments in large firms and all other large private
sector establishments every year. If a larger amount of money were available, a third option
for augmenting the OES sample would be to combine the first two options. Under this option,
data would be collected annually for all establishments that belong to firms with 10,000 or more
employees and for all other establishments with 250 or more employees. To estimate the cost of
this option, we assume that the incremental costs of data collection for establishments belonging
to large firms would be $8,000–$10,000 per firm and that collecting data for any remaining
establishments with 250 or more employees would cost $100–$200 per establishment, depending
80
on establishment size, the same assumptions used to estimate the costs of option 1 and option 2.
At $8,000 per firm for COC collections, this option would raise the cost of the OES program by
$6.4 million dollars per year; if we instead assume a cost of $10,000 per firm for COC
collections, the cost increment would be $7.7 million. These figures are about $1.0 million less
than the sum of the incremental costs for option 1 and option 2, due to the overlap between the
two covered groups, consisting of large establishments that belong to large firms.
Option 3 would reduce the variance of annual OES estimates by a larger amount than either
option 1 or option 2. In total, it would add establishments with 27.8 million jobs to the sample
each year. Firm-level analysis also would be supported under this sample design option, at least
for the set of large firms for which comprehensive data were collected.
Other options. The options outlined above are of course not an exhaustive set of possibilities.
One might, for example, want to modify option 1 by collecting data each year not only for large
establishments but also for a subsample of smaller establishments. Option 2 might be modified
by focusing collection efforts on firms doing business primarily in sectors that, according to
some yet-to-be specified criterion, are likely to be affected by offshoring activity. Even
assuming a cost of no more than $8,000–$10,000 per firm, reductions in the firm size threshold
for data collection that were applied economy-wide would be very expensive. For example,
lowering the firm size threshold to include all firms with employment of 5,000 or more would
raise the cost of option 2 to between $10.5 million and $13.8 million. A lower firm size
threshold might be more feasible, however, if its application were restricted to certain sectors.
Other variants of the sampling options we have outlined also could be devised, but all would
have in common the designation of some significant sample of establishments for annual
collection.
One question that needs to be asked about the survey redesign options we have suggested is
whether businesses would in fact be willing to respond to the survey every year. The results of a
test conducted in 2000 are at least somewhat encouraging. In this test, staff in 25 states
attempted to collect data from all of the certainty units in their states, and 12 states provided data
on their collection experience that could be analyzed. These data show roughly comparable
response rates for the certainty units originally scheduled to participate in the OES in 2000 and
the added units originally scheduled to report only in another year of the three-year collection
cycle. On the negative side, collecting data from both groups was difficult and the combined
response rate was lower than would be hoped. An important factor in the response rate obtained
appears to have been the amount of staff time available for follow-up with the surveyed units
(BLS 2001b).
Another question is whether large companies would be willing to submit electronic records for
all of their establishments centrally, as is assumed in our cost estimates for option 2 and option 3.
Some large companies already do this and, depending on how their records are kept, others
might be willing to do so. But if a significant number of large companies cannot or will not
agree to central office collection, response rates for options that envision such collection could
be adversely affected and the data collection costs for these options could be substantially higher
than our rough figures suggest.
81
Other Changes in Survey Management Practices and Philosophy
In addition to the changes in sample design discussed above, converting the OES from a survey
designed only to produce detailed cross-sectional estimates to a survey that also produced usable
time series would require a number of other changes to the way in which the survey was
managed. These would include changes in editing and imputation procedures, plans for dealing
with future changes in the industry and occupation classification structures, and overarching
changes in the philosophy governing other management decisions. Making these changes almost
certainly would require additional staffing, and it should be emphasized that the costs of this
added staffing are not reflected in the data collection cost estimates cited above.
Consider the process for editing the survey responses that are submitted. Under the current
program structure, responses submitted by establishments are reviewed in isolation. Edit checks
examine whether the combination of occupations reported seems sensible according to a
specified set of criteria, but there are no edit checks that examine whether the information
reported by an establishment in the current year is consistent with that reported by the same
establishment in previous years. Checking the consistency of establishment reports across years
would be more important in a program designed to produce time-series data. For example,
inconsistencies in occupational coding across years might be identified through edit checks that
flagged shifts of large blocks of employment from one occupation to another.
Similarly, under the current survey design, data for establishments that do not respond are
imputed using information for other establishments that are deemed to be similar. In a program
designed to produce time-series data, however, given idiosyncratic variation in staffing patterns
across establishments, it would be desirable to develop an imputation methodology that relied
more heavily on data reported by the missing establishments in previous time periods.
A recurring issue for a program designed to produce time-series estimates of employment by
occupation would be dealing with future changes to the SOC and NAICS. The next scheduled
NAICS revision is set for 2017, and the next scheduled SOC revision for 2018. From that point
forward, current plans call for NAICS revisions approximately once every 5 years and SOC
revisions approximately once every 10 years. It seems likely that SOC revisions will pose the
most serious challenges for the OES program. Having a set of dual-coded records containing
both the old and the new SOC code could allow the OES staff to reconstruct historical
occupational employment series on the new SOC basis, but dual coding is expensive and would
need to be built in to the OES budget plans.
Less tangible, OES management and staff would need to reorient themselves toward a new set of
survey objectives, which in turn would drive subsequent decisions. There is always a tension
between making changes designed to improve survey estimates and preserving the continuity of
historical series. In the current OES program, the program’s stated objectives have dictated that,
when data improvements are possible, they should be introduced, even when that makes the data
less comparable over time. In a program that had as one of its stated objectives the production of
annual time series, this balance would need to be set differently.
82
CONCLUSION
Because the annual sample for the OES survey is large, it is reasonable to think about using the
historical OES data to produce annual time-series estimates. In practice, however, while useful
for certain purposes, the annual times series that the historical data will support have significant
limitations for studying the effects of offshoring. Reflecting the fact that certainty units are
surveyed only once every three years, annual estimates for detailed occupations can be expected
to have high variance. Breaks in occupation and industry classification systems and other
changes in survey practices further complicate any trend analysis based on the OES data.
Further, the existing data are not well suited to support longitudinal analysis.
Any redesign of the OES program to produce reliable annual estimates and/or support
longitudinal analysis should address each of these factors. We have suggested several options
for redesigning the survey sample that would involve the collection of data from all large private
sector establishments each year and/or the collection of data from all of the establishments at
selected large firms each year. Ballpark cost estimates for the sample expansions associated with
these options range from $3.5 to $7.7 million per year, though we should emphasize both that
these estimates are very rough and that they cover only the direct costs of data collection. If the
OES survey were to be redesigned along the lines we have suggested, funding also would be
required to support new data editing procedures, dual coding of survey records at the time of
future changes to the SOC, and other survey management activities, though data collection likely
would account for the largest share of the total new funding that would be needed.
A significant complication we have not addressed is whether and how annual published
estimates would be reconciled with more detailed cross-sectional estimates produced using data
from multiple waves of data collection. The existing OES customer base cares a great deal about
geographically disaggregated estimates and our guess is that, even with an expanded sample, the
needs of this customer base could not be satisfied by purely annual data. This is, however, at
least partly an empirical matter that remains to be addressed. Even with the full sample of
approximately 1,200,000 establishments currently collected over three years, the BLS does not
publish occupational data for industries disaggregated by geographic area, and, if we have
understood correctly what is being recommended, it seems unrealistic to call for “. . . the BLS to
make the changes to the OES methodology necessary to create time series data on all 820
occupations in the SOC by industry and geographic areas” (Sturgeon 2006). Further work would
be required to determine the level of detail in occupational employment estimates—whether by
industry or by geographic area, but almost certainly not by both simultaneously—that could be
supported by different sample redesign options.
Finally, in thinking about a possible OES redesign, it will be important to consider carefully the
value of data to support firm-level longitudinal analysis as compared to the value of improved
annual time series. Without a doubt, a longitudinal occupational employment database could
support interesting research, including useful research on the effects of offshoring. Further, there
may be significant economies associated with the collection of data from all establishments of
large firms. On the other hand, adding establishments from a small number of firms can be
expected to do less to reduce the variability of annual time-series estimates than adding a similar
number of establishments representing a larger number of firms. In addition, unless new
modalities for researcher access to confidential microdata are developed, the number of
83
researchers who would in practice end up working with firm-level data from the OES seems
likely to be limited. Those charged with making a decision about the survey’s future will need to
be clear about the relative importance of different survey objectives as they choose among
possible redesign options.
84
Figure 1: OES Employment of Ten Occupations Identified
by Jensen and Kletzer (forthcoming)
As Most Off-Shorable, 1999-2004
191021: Biochemists & Biophysicists
152091: Mathematical Technicians
2,000
20,000
1,000
15,000
0
10,000
1999
2000
2001
Published
2002
2003
2004
1999
Annual Microdata
2000
2001
Published
152041: Statisticians
2002
2003
2004
Annual Microdata
232093: Title Examiners, Abstractors, ...
25,000
70,000
15,000
50,000
5,000
30,000
1999
2000
2001
Published
2002
2003
2004
1999
Annual Microdata
2000
2001
Published
2002
2003
434041: Credit Authorizers, Checkers, & Clerks
90,000
435111: Weighers, Measurers, Checkers, ...
95,000
70,000
80,000
50,000
2004
Annual Microdata
65,000
1999
2000
2001
Published
2002
2003
1999
2004
Annual Microdata
2000
2001
Published
439021: Data Entry Keyers
2002
2003
2004
Annual Microdata
132011: Accountants & Auditors
550,000
1,100,000
400,000
900,000
250,000
700,000
1999
2000
2001
Published
2002
2003
2004
1999
Annual Microdata
2000
2001
Published
319094: Medical Transcriptionists
2002
2003
2004
Annual Microdata
152011: Actuaries
100,000
24,000
85,000
17,000
70,000
10,000
1999
2000
2001
Published
2002
2003
2004
1999
Annual Microdata
2000
2001
Published
85
2002
2003
2004
Annual Microdata
References
Abraham, Katharine G., and James R. Spletzer. Forthcoming. “Are the New Jobs Good Jobs?”
In Labor in the New Economy, Katharine G. Abraham, James R. Spletzer, and Michael
Harper, eds. Chicago: University of Chicago Press.
Bureau of Labor Statistics (BLS). 2001a. Occupational Employment and Wages, 1999, Bulletin
2545. Washington, DC: Government Printing Office. September.
http://www.bls.gov/oes/1999/oes_pub99.htm (accessed October 10, 2009).
———. 2001b. Analysis of Certainty Unit Masterfiles, 2000 OES Survey. Certainty Unit
Subgroup. OES Policy Council Meeting, October 10–11, Boston, MA.
———. 2008. Report on Current Costs, Overhead and Response Measures. NCS-OES Data
Collection and Processing Cost Team. Unpublished report. January 8. Washington, DC.
Jensen, J. Bradford, and Lori G. Kletzer. Forthcoming. “Measuring Tradeable Services and the
Task Content of Offshorable Services Jobs.” In Labor in the New Economy, Katharine
G. Abraham, James R. Spletzer, and Michael Harper, eds. Chicago: University of
Chicago Press.
Mikkelson, Gordon, Teresa L. Morisi, and George Stamas. 2000. “Implementing the NAICS for
Business Surveys at BLS.” Paper presented at the Second International Conference on
Establishment Surveys, held in Buffalo, NY, June 17–21.
National Academy of Public Administration. 2006. Off-Shoring: How Big Is It? Report to the
U.S. Congress and the Bureau of Economic Analysis, Washington, DC.
Office of Management and Budget (OMB). 2000. Standard Occupational Classification
Manual. White Plains, MD: Graphic Systems, Inc.
Sturgeon, Timothy J. 2006. Why We Can’t Measure the Economic Effects of Services
Offshoring: The Data Gaps and How to Fill Them. Final report of the Services
Offshoring Working Group, Industrial Performance Center, Massachusetts Institute of
Technology, Cambridge, MA.
87
Understanding the Domestic Labor Market Impact of
Offshore Services Outsourcing: Measurement Issues
Lori G. Kletzer
University of California, Santa Cruz and
Peterson Institute for International Economics
October 29, 2009
Lillian Wilson provided excellent research assistance. Financial support was provided by the
Labor and Employment Research Fund in the University of California Office of the President.
Correspondence: <[email protected]>
89
Over a remarkably short time frame, thinking about the production of services and services
employment has changed dramatically. For a sector once considerable “nontradable,” measuring
services trade is now a task of considerable energy and importance. The possible domestic
employment implications of rising services trade—that is, services offshoring—attract
significant attention and political interest.31 The literature on services offshoring is expanding,
although the activity remains “an elusive phenomenon.”32 Much of that elusiveness springs from
data limitations and measurement concerns, the subject of this conference. The increasing role of
multinational corporations (MNCs), technology transfer, and increasing trade in services are just
three activities associated with globalization where statistical and measurement limitations are
widely acknowledged. Scholars interested in the labor market implications of increasing trade in
services add one more area to this (partial) list: detailed time-series occupational data.
Understanding the relocation of work, but not the workers doing the work, and more broadly,
understanding the nature of work that is potentially offshorable, requires consistent detailed
occupational data.
Before turning to the current data limitations and ongoing efforts to address the limitations, it is
useful to establish definitions and provide context. For the vein of research focusing on domestic
employment, one research goal is to understand and measure the magnitude and significance of
shifting of business operations to offshore (foreign) locations and impact on American workforce
(employment and earnings). The labor market impact questions include estimates of jobs moved
offshore; estimates of lost potential job growth (because jobs moved offshore); jobs added due to
foreign work located in the United States; jobs added due to efficiency gains; shifts in
occupations; changes in earnings and the distribution of earnings; job displacement (numbers,
worker characteristics, unemployment durations, UI receipt, reemployment).
Stated this way, it becomes clearer that the offshoring questions are not new, in that similar
questions have been asked about and for the manufacturing sector for decades. These questions
have, however, taken on heightened awareness with growth of services trade. This heightened
awareness arises in part from the manufacturing/production worker focus becoming business and
professional services/“white-collar” and professionals. This does not imply that everything is the
same and nothing is new. Rather, the potential for services offshoring highlights tasks and
occupations in a way that manufacturing import competition did not (and does not). Also, the
implications for educational attainment may be different, as a result of the different impact on
occupations.
One focused avenue in the literature on services offshoring attempts to address directly the
occupational or task nature of the activity. Papers by Jensen and Kletzer (2006, 2008); Blinder
(2006, 2007); van Welsum and Reif (2009); and Moncarz, Wolf, and Wright (2008) share a
general approach to measuring potential offshorability by looking at the task and activity content
of jobs. Although these papers differ in methodology and details, they share a common starting
31
Services offshoring refers to the (potential) migration of jobs (but not the people performing them) across national
borders, mostly from rich countries to poor ones, with imported products and activities flowing back to the United
States.
32
See Jensen and Kletzer (2008) for citations to the literature. The phrase “elusive phenomenon” appears in NAPA
(2006a). Interested readers are directed to the set of three NAPA publications on offshoring (2006a,b; 2007).
91
point that movable jobs are those with little face-to-face customer contact; high information
content, work process is Internet enabled and/or telecommutable (see Bardhan and Kroll [2003];
Dossani and Kenney [2003], and Blinder [2006]). More informally, it is commonly believed that
if “it can be sent down a wire (or wireless),” it is offshorable. These papers have all yielded sets
of occupations varying in their “potential offshorability.” One possible next step, as noted in
Chapter 4 of NAPA (2006b), is to consider that the offshoring of services should produce
changes in the occupational structure of firms and establishments. In other words, shifts in
certain (potentially “movable”) occupations may be consistent with offshoring. Shifts in these
occupations, within industries with (intermediate) services trade, may be more compelling
evidence.
This paper proceeds with some thoughts on possible enhanced links between studies of
(domestic) outsourcing and (international) offshore outsourcing. After that I describe the basic
principles of offshorability and the data on the content and context of jobs (O*Net), together with
occupational employment and earnings (Occupational Employment Statistics [OES]). The next
section considers the preliminary evidence of shifts of the occupational employment distribution,
followed by a concluding section.
HOW STUDIES OF OFFSHORING MAY BENEFIT FROM LITERATURE ON
OUTSOURCING
Over the past five years, the potential for services offshoring has generated remarkable attention
for an internal-to-the-firm economic activity, and area of research, historically undertaken
behind-the-scenes and not in the spotlight. These behind-the-scenes activities are the “make-orbuy” points—the decisions to use in-house (own) employees versus purchasing completed tasks
from other establishments.
Interest in domestic contracting out surged in the 1990s, with attention paid to temporary help
agencies. Katharine Abraham’s research led the way, with her 1988 and 1990 papers on marketmediated work arrangements. Then, as now, research progress was slowed by data limitations.
The still-small literature on domestic outsourcing now reveals significant growth in the activity
over the past 25 years (see papers by Segal and Sullivan [1997], Houseman [2001], and Dey,
Houseman, and Polivka [forthcoming]).
To date, research on offshore outsourcing has proceeded without much of a link to the domestic
outsourcing literature. Interestingly, perspectives from international trade have dominated the
offshore outsourcing literature, in a services version of “does trade cost jobs?” Yet the domestic
outsourcing literature has implications for offshore outsourcing research. Domestic outsourcing
can produce shifts in the industry structure of employment. Contractual production workers are
employees of temporary help agencies (or more broadly, firms in the business services sector).
In-house production workers are (usually) employed by manufacturing firms. Simply put, as
stated by Dey, Houseman, and Polivka (forthcoming), “. . . the number and occupational
distribution of workers classified in the manufacturing sector changes, even if the number and
occupational distribution of workers performing the tasks does not” (p. 2). Thus, the domestic
outsourcing literature looks at shifts in the pattern of occupational employment across
92
industries.33 Following the logic that offshore outsourcing involves the reallocation of
production tasks within establishments, firms, and industries, we might expect to see shifts in the
pattern of occupations within industries, and most specifically for “potentially movable”
occupations within industries where business and professional services imports have increased.
Investigating this (weakly formed) hypothesis involves bringing together data on (potentially
movable) occupations, occupational employment, and services trade.
MEASURING TASK CONTENT OF POTENTIALLY MOVABLE SERVICES
OCCUPATIONS34
The literature on offshoring posits that movable jobs are those with little face-to-face customer
contact, those with high information content, and those whose work processes are Internet
enabled and/or telecommutable.35 A great deal of attention is paid to Internet-enabled: the
expansion of broadband and wireless (and the broad use of “off the shelf” software programs)
having greatly reduced the “transportation costs” of information. Having developed a set of
tradable services occupations, the next step is to consider the detailed characteristics of these
jobs and whether the characteristics fit a description of offshorability.
The use here of Occupational Information Network (O*Net) is in the spirit of Autor, Levy, and
Murnane (2003), who explore the spread of computerization, using the Dictionary of
Occupational Titles (DOT) to measure the routine vs. nonroutine, and cognitive vs. noncognitive
aspects of occupations. The O*Net was developed by the U.S. Department of Labor as a
replacement for the DOT.36 Similar in theme to the DOT as a source of occupational information,
O*Net reflects the expanded possibilities of contemporary information technology in that it is a
database, with information on job characteristics and worker attributes. Unlike the vast jobspecific detail provided on 12,000+ occupations in the DOT, O*Net provides information on
1100+ occupations, using language and assessment common across jobs. Unlike DOT, where
professional analysts are the primary source of information, job incumbents provide the
information, which is gathered by survey questionnaire. Occupations are organized at the
Standard Occupational Classification level.37 O*Net is used in a variety of fields studying work
and occupations, such as organizational behavior, applied psychology, career assessment, human
resource management, and occupational psychology.38 O*Net is relatively foreign to research in
economics. Blinder (2007) takes an approach similar in spirit to our discussion here.
33
In fairness to the domestic outsourcing literature, it considers a broader range of questions from industrial
relations to labor demand, including job security, wages, compensation and benefits costs, job training, hiring,
firing, and search costs.
34
This section borrows heavily from Jensen and Kletzer (2008).
35
See Bardhan and Kroll (2003) for a list of attributes.
36
See Peterson and Mumford et al. (1999) for a history of the development of O*Net.
37
Importantly, the level of SOC detail used in O*Net (6-digit plus) is deeper than the 6-digit SOC codes used in
OES.
38
See http://online.onetcenter.org/ for information on acquiring the data.
93
The O*Net Content model identifies the most important types of information about work, jobs,
and workers, and integrates the information into a structured system of six major categories:39
1) Worker characteristics (abilities, occupational interests, work values, work styles);
2) Worker requirements (skills and knowledge, education);
3) Experience requirements (experience and training, skills and entry requirements,
licensing);
4) Occupational requirements (generalized and detailed work activities, organizational
context, work context)
5) Labor market characteristics (labor market information, occupational outlook); and
6) Occupation-specific information (tasks, tools and technology).
The first three categories (Worker characteristics, Worker requirements, Experience
requirements) are worker oriented. The second three are work- (or job-) oriented categories, with
Occupational requirements as the focus of interest here. Occupational requirements are meant to
identify requisite tasks, and are designed to cross occupations, at both a general and detailed
level, while Occupation-specific information is meant to be quite detailed and literally
occupation specific.
The domain/category Occupational requirements is designed to provide “. . . a comprehensive set
of variables or detailed elements that describe what various occupations require” (National
Center for O*Net Development 2006, p. 20). The focus is on typical activities required across
occupations. Within the Generalized and detailed work activities subdomain, 11 measures to
construct an index of offshorability/potential tradability. The sign in parentheses [(+) or (-)]
denotes a prior on whether the characteristic is positively related to offshorability or negatively
related.
On information content:
Getting information (+)
Processing information (+)
Analyzing data or information (+)
Documenting/recording information (+)
On Internet-enabled:
Interacting with computers (+)
On face-to-face contact:
39
The idea behind the six content areas is to provide “multiple windows” on the world of work. Information on the
O*Net Context Model comes from National Center for O*Net Development (2006). For a comprehensive discussion
of O*Net from the practical and research perspectives, see Peterson and Mumford et, al. (2001).
94
Assisting or caring for others (-)
Performing or working directly with the public (-)
Establishing or maintaining interpersonal relationships (-)
On working together or supervising the work of others:
Communicating with supervisors, peers, subordinates (-)
Training and teaching others (-)
Performing administrative duties (-)
Coordinating work and activities of others (-)
On the “on-site” nature of work:
Inspecting equipment, structures, or material (-)
Monitoring processes, materials, and/or surroundings (-)
Rating scales are used to quantify these characteristics. Multiple scales are provided, with
“importance” and “level” as the predominant pair. “Importance” is the rating of answers to the
question, “How important is this skill to performance on the job?” Answers vary from “not
important” to “extremely important,” on a scale of 1 to 5. “Level” is the rating of “What level of
this skill is needed to perform this job?” ranging from low (level) to high (level), on a scale of 1
to 7.40 An illustration might be useful, normalizing the two different scale ranges to 0 to 100. For
the attribute “Performing or working directly with the public,” data entry keyers are assigned
importance (I) =43, and level (L) = 33. For Security Guards, I=74 and L=62. Compared to data
entry keyers, working with the public is more important to performance on the job for security
guards, along with a higher level of the “skill” of working with the public. See Jensen and
Kletzer (2008) for a more complete description of the rating scheme and a presentation of
summary statistics on the work activities.
The composite index of offshorability is the weighted sum of the 14 components, using priors on
the sign of the attribute in regard to offshoring potential. Higher values of the index indicate
more offshorability potential, yielding a ranking of all occupations for which the attributes are
available.41 The usefulness of the index is ordinal, not cardinal. Occupations are judged on their
offshorability relative to each other, not compared to some absolute standard. Tables 1 and 2
report the top-30 and bottom-30 occupations.42 Occupations at the top of the list seem
unsurprising: credit authorizers, data entry keyers, accountants, medical transcriptionists, market
research analysts, bookkeeping, and account clerks.
40
See Peterson and Mumford et al. (1999, 2001). Level allows a “not relevant to performance” rating, coded as 0.
In constructing an index, it is not obvious how to weight importance and level. Starting from the observation that
importance varies more than level across occupations, an index was created using a weight of three-quarters to
importance and one-quarter to level. The ranking is robust to different weights.
42
The full listing of 799 occupations, ranked by job-task content, takes up 28 printed pages, and is available upon
request.
41
95
Tables 1 and 2 list employment and median annual earnings for each occupation, for May 2003
and May 2008, obtained from the OES program. The OES program, operated by the U.S. Bureau
of Labor Statistics (BLS), generates employment and earnings estimates for over 800 detailed
occupations, derived from a semiannual mail survey of establishments. Although the OES survey
methodology is designed to create detailed cross-sectional employment and wage estimates for
the U.S. and smaller geographic units, across and by industry, it is less useful for comparisons of
two or more points in time. Changes in the procedures for collecting data, along with changes in
occupational and industry codes may it complicated to create a time series. A great deal of detail
must be suppressed to create a consistent time series, as noted by Abraham and Spletzer (2009).
Dey, Houseman, and Polivka (forthcoming) create a time series for 15–18 broad occupational
aggregates (at the major occupation level) and 6 narrow occupational groups. This level of
aggregation loses the movability characteristics available from O*Net. For this paper’s
preliminary analysis, changes in employment and earnings are considered at just two data points,
May 2003 and May 2008.
To date, I have located three other analyses that order occupations by an assessment of
offshorability. Consistent with its organizational interest in occupational growth projections, the
BLS has developed a list of 40 detailed occupations deemed “susceptible to a significant risk of
offshoring” (BLS 2006, p. 12). Of these 40 occupations, 39 are services occupations (the
exception is aircraft mechanics and service technicians). With varying degrees of “fit,” 38 of
these 39 occupations are noted for their offshorability by the index reported here. Graphic
designers and switchboard operators are included in the BLS list, with my index ranking these
two occupations close to the middle of the 457. All the rest of the BLS occupations are fairly
highly ranked by my index. The BLS list is not ranked; it is simply offered as a list of susceptible
occupations, presumably with some more susceptible than others.43
Moncarz, Wolf, and Wright (2008) present a more comprehensive analysis of offshoring and
occupations, from work performed for the BLS Employment Projections Program. Starting with
515 service-providing occupations, BLS economists who study occupations identified those
occupations “that had insurmountable barriers to offshoring” (p. 73).44 After eliminating
occupations “considered not at all susceptible to offshoring” (p. 73), the analysis was confined to
160 occupations. The analysts considered four characteristics: 1) inputs and outputs that can
travel easily across long distances (such as electronically), 2) work that requires little interaction
with other types of workers, 3) work that requires little knowledge of the social or cultural
idiosyncrasies of the target market, and 4) work that is routine in nature (p. 73). Occupations
were scored on this characteristics (very low to very high, a four-point scale), and assigned a
susceptibility score. A preliminary comparison of the resulting ranking suggests considerable
difference between the BLS analysis and the analysis reported here.
43
The BLS methodology is similar in spirit to ours, considering characteristics of digital transmission, repetitive
tasks, little face-to-face interaction. Occupational analysts provided judgments on these characteristics. Further
refinements included excluding occupations where technology or automation could account for a dampening of
employment growth. See BLS (2006).
44
Examples include physical therapists and barbers, security guards, correctional officers.
96
Blinder (2007) explores a subjective index based on two characteristics: 1) can the work be
delivered to a remote location, and 2) must the job be performed at a specific (U.S.) location? In
his subjective measure, Blinder concentrates on one characteristic of the delivery of services, the
separation of customer and supplier that he labels “impersonally delivered services.” Basically,
impersonally delivered services can be delivered electronically, incorporating the vast
improvement in ICT. His measure does not incorporate any attributes related to the kind of work
sent down the wire, such as information context or internet enabling. Most importantly, in terms
of the area of traditional US comparative advantage, Blinder does not consider the creativity or
routineness of work.45 In an area that needs more exploration, there are many high-skill and highvalue (creative) services, that while transmittable electronically, pose opportunities for American
workers and firms to penetrate foreign markets.
Using both production and nonproduction occupations, Blinder estimates that 30–40 million
workers are currently in potentially tradable jobs, based on May 2005 employment levels.
Objective measures may well be preferred, given the number of occupations (less than 450) and
desire for replication.
RISING SERVICES TRADE AND SHIFTS IN THE OCCUPATIONAL DISTRIBUTION
OF EMPLOYMENT
Taking up the approach followed in National Academy of Public Administration (NAPA 2006b),
this section examines shifts in the occupational distribution of employment within service sector
industries where imports (trade) have expanded. The idea is to look for evidence consistent with
offshoring, within industries where services trade has expanded: Do high potential movability
occupations decline (relative to national trends) in industries coincident with rising imports? 46
How does the occupational employment share within a “rising import” industry relate to
potential movability?
NAPA (2006b) examines a limited set of industries that were “significant in size, potentially
vulnerable to off-shoring, sufficiently diverse, well integrated into the overall economy, and
likely to continue expanding.” Four industries were selected:
1) pharmaceutical and medicine manufacturing [3254];
2) architectural, engineering, and related services [5413];
3) computer systems design and related services [5415]; and
4) business support services [5614].
The NAPA analysis developed additional and more extensive measures of offshoring than
tackled here (to date). The approach here is in the spirit of NAPA’s analysis, of considering
service industries with a high-tech component with increasing trade flows, yet with a
considerably larger set of detailed industries. Appendix Table A.1 reports changes in
45
The routineness of work, or the codification of tasks, is a characteristic emphasized by Autor, Levy, and Murnane
(2003).
46
This framing ignores the question of whether the appropriate “trade” measure is the level of trade or the change.
97
occupational employment, within 6-digit NAICS industries and across industries, for NAICS
sectors 51 (Information), 54 (Professional, scientific, and technical services) and 56
(Administrative and support services).47
Before turning to the occupational employment data, Figure 1 shows the rise in trade in the
overall category of Other private services, where these NAICS sectors reside. There is a trade
surplus in this category that has grown since the early 2000s. Table 3 presents more detailed
trade data, for the period 2002–2007, for a subset of Other private services that includes a
number of the industries examined here. The services trade surplus is broadly in evidence,
although it is also clear that imports have increased substantially.
Returning to Appendix Table A.1, for each industry, the subset of occupations charted starts with
the “Ten largest occupations for each industry” featured on the BLS Web site, drawn from the
May 2008 OES (BLS 2009). Given the difficulties of using the OES data as a time series, and the
desire to examine very detailed occupations, the analysis to date compares just two points in
time, May 2003 and May 2008.
Table 1.1 is sufficiently detailed to make summary statistics complicated. Offering the most
summary of summary statistics, the movability index is negatively correlated with changes in the
share of industry employment accounted for by an occupation (if an occupation’s share of
industry employment rose from 2003 to 2008, that occupation was lower ranked in terms of
potential movability).48 A general observation from the table: the majority of the most populous
occupations in these industries grew faster than the corresponding national occupational average
(that is, a comparison of Columns 2 and 4).
CONCLUSIONS
This paper offers a measure of potential movability of occupations, built from common notions
of job characteristics related to “offshorability.” The calculated index of offshorability offers
strong potential for understanding jobs (tasks) at risk. Preliminary analysis of how these
occupations have changed in importance, measured as by employment share and earnings
change, is ongoing. A natural question arises as to whether business, professional, and technical
services industries with rising imports show evidence of shifts in occupational employment that
are consistent with offshoring.
47
NAICS sector 56 also includes Waste management and Remediation services, but those detailed industries are not
included here.
48
Correlation = –0.25 for sectors 54 and 56.
98
Table 1: Top 30 Occupations for Potential Movability
May 2008
SOC code Occupation title
Movability
index Rank
Employment
May 2003
Median
annual
earnings
Employment
Change, 2003–2008
Median
annual
earnings Employment
Real
median
earnings
15-2091
Mathematical technicians
1.274
1
1,100
$38,400
2,180
$36,540
-0.495
-0.102
15-2021
Mathematicians
0.118
2
2,770
$95,150
2,470
$78,290
0.121
0.039
17-3013
Mechanical drafters
-0.171
3
77,070
$46,640
74,010
$41,520
0.041
-0.040
13-2041
Credit analysts
-0.173
4
74,400
$55,250
68,910
$45,020
0.080
0.049
15-2031
Operations research
analysts
-0.289
5
60,860
$69,000
58,080
$58,300
0.048
0.012
19-2011
Astronomers
-0.358
6
1,280
$101,300
770
$88,310
0.662
-0.020
15-2041
Statisticians
-0.391
7
20,680
$72,610
18,370
$59,560
0.126
0.042
17-3012
Electrical and electronics
drafters
-0.395
8
32,710
$51,320
33,720
$41,730
-0.030
0.051
15-1051
Computer systems analysts
-0.443
9
489,890
$75,500
474,780
$64,160
0.032
0.006
19-3011
Economists
-0.454
10
12,600
$83,590
12,300
$70,250
0.024
0.017
19-3021
Market research analysts
-0.457
11
230,070
$61,070
142,190
$54,670
0.618
-0.045
15-2011
Actuaries
-0.486
13
18,220
$84,810
14,680
$72,520
0.241
0.000
15-1031
Computer software
engineers, applications
-0.506
14
494,160
$85,430
392,140
$72,530
0.260
0.007
17-2072
Electronics engineers,
except computer
-0.539
15
139,930
$86,370
137,320
$71,370
0.019
0.034
13-2031
Budget analysts
-0.564
16
62,630
$65,320
55,560
$54,520
0.127
0.024
15-1061
Database administrators
-0.571
17
115,770
$69,740
100,890
$58,200
0.147
0.024
13-2011
Accountants and auditors
-0.577
18
1,133,580
$59,430
924,640
$49,060
0.226
0.035
43-9021
Data entry keyers
-0.644
19
272,810
$26,120
339,010
$22,600
-0.195
-0.012
13-2052
Personal financial advisors
-0.668
20
146,690
$69,050
85,670
$58,700
0.712
0.005
23-2093
Title examiners,
abstractors, and searchers
-0.669
21
59,390
$38,300
47,840
$34,080
0.241
-0.039
17-2071
Electrical engineers
-0.671
22
154,670
$82,160
146,150
$69,640
0.058
0.008
31-9094
Medical transcriptionists
-0.677
23
86,200
$32,060
97,810
$27,590
-0.119
-0.007
17-2011
Aerospace engineers
-0.679
24
67,800
$92,520
70,740
$74,520
-0.042
0.061
17-1021
Cartographers and
photogrammetrists
-0.741
25
11,690
$51,180
8,940
$44,170
0.308
-0.010
19-3041
Sociologists
-0.754
26
4,390
$68,570
3,060
$54,410
0.435
0.077
43-9111
Statistical assistants
-0.760
27
16,900
$34,850
20,970
$29,890
-0.194
-0.003
43-3031
Bookkeeping, accounting,
and auditing clerks
-0.763
28
1,855,010
$32,510
1,750,680
$27,760
0.060
0.001
15-1011
Computer and informa-tion
scientists, research
-0.785
29
26,610
$97,970
23,210
$81,600
0.146
0.026
13-2051
Financial analysts
-0.833
30
236,720
$73,150
165,420
$60,050
0.431
0.041
99
Table 2: Bottom 30 Occupations for Potential Movability
May 2008
SOC
code Occupation title
31-9091 Dental assistants
Movability
Index Rank Employment
-3.228 768
293,090
May 2003
Median
annual
earnings
$32,380
53-6021 Parking lot attendants
-3.241 769
136,470
$18,790
29-1061 Anesthesiologists
-3.247 770
34,230
#
39-5093 Shampooers
-3.257 771
15,570
29-1081 Podiatrists
Change, 2003 -2008
Median
Real
annual
median
Employment earnings Employment earnings
272,030 $27,700
0.077
-0.001
113,490 $16,630
-0.034
#
0.439
$17,300
15,300 $14,360
0.018
0.030
-3.272 772
9,670 $113,560
7,800 $94,060
0.240
0.032
47-5061 Roof bolters, mining
-3.284 773
4,950
$45,210
3,980 $38,550
0.244
0.002
25-2011 Preschool teachers, except special
education
-3.288 774
392,170
$23,870
368,870 $19,820
0.063
0.029
27-2032 Choreographers
-3.301 775
13,860
$38,520
14,810 $31,030
-0.064
0.061
29-9091 Athletic trainers
-3.316 777
15,070
$39,640
11,750 $32,850
0.283
0.031
29-2055 Surgical technologists
-3.317 778
89,600
$38,740
73,250 $32,130
0.223
0.031
53-3011 Ambulance drivers and attendants,
except emergency medical technicians
-3.324 779
21,790
$22,410
17,650 $19,000
0.235
0.008
33-9032 Security guards
-3.334 780
1,046,760
$23,460
964,260 $19,660
0.086
0.020
39-1021 First-line supervisors/managers of
personal service workers
-3.340 781
129,070
$34,910
110,630 $29,500
0.167
0.011
21-2011 Clergy
-3.351 782
42,040
$41,730
38,170 $33,800
0.101
0.055
27-2021 Athletes and sports competitors
-3.362 783
13,960
$40,480
11,840 $45,780
0.179
-0.244
39-3091 Amusement and recreation attendants
-3.366 784
258,820
$17,470
236,070 $15,030
0.096
-0.007
31-9011 Massage therapists
-3.379 785
51,250
$34,900
29,940 $28,670
0.712
0.040
39-3011 Gaming dealers
-3.381 786
91,130
$16,310
76,120 $14,200
0.197
-0.018
49-9051 Electrical power-line installers and
repairers
-3.427 787
111,580
$55,100
95,190 $48,960
0.172
-0.038
47-2022 Stonemasons
-3.446 788
18,910
$37,800
13,710 $34,000
0.379
-0.050
47-4091 Segmental pavers
-3.480 789
1,170
$27,400
1,710 $26,530
-0.316
-0.117
27-2041 Music directors and composers
-3.488 790
9,120
$41,270
9,000 $32,530
0.013
0.084
37-3013 Tree trimmers and pruners
-3.497 791
35,420
$29,970
40,710 $25,630
-0.130
-0.001
11-9031 Education administrators, preschool and
child care center/program
-3.528 792
49,630
$39,940
56,030 $34,500
-0.114
-0.011
29-2054 Respiratory therapy technicians
-3.533 793
16,210
$42,430
25,470 $34,850
-0.364
0.041
39-6031 Flight attendants
-3.609 794
99,480
$35,930
107,100
***
-0.071
39-9031 Fitness trainers and aerobics instructors
-3.623 796
229,030
$29,210
177,790 $24,510
0.288
0.019
33-1021 First-line supervisors/managers of fire
fighting and prevention workers
-3.680 797
53,300
$67,440
59,000 $57,000
-0.097
0.011
27-2022 Coaches and scouts
-3.844 798
175,720
$28,340
105,070 $26,950
0.672
-0.101
49-9095 Manufactured building and mobile
home installers
-4.131 799
8,290
$28,250
13,160 $23,360
-0.370
0.034
100
23,790
0.202
Table 3: International Trade in Business, Professional and Technical Services (millions of dollars)
2002
2003
2004
2005
2006
2007
Total
Imports
34185
37458
40992
46924
61068
68763
Exports
60177
62958
69568
76487
89692
107675
Imports
786
864
931
876
1845
1977
Exports
466
517
581
896
3163
4030
Imports
6495
7617
8639
10596
13085
14815
Exports
7079
8213
8693
9434
10341
12798
Imports
4063
5071
5778
7239
9429
11437
Exports
8678
9467
9563
10431
12821
14698
11028
10770
12076
14905
19361
20475
14339
14309
16372
19242
22058
24699
Imports
820
874
899
894
1222
1561
Exports
3099
3377
3997
4225
5294
6424
316
303
580
434
1751
1851
2247
2564
3294
3791
5369
6469
Imports
183
176
164
169
1035
1504
Exports
806
877
828
2303
3836
3872
Imports
668
670
720
956
3780
4180
Exports
5287
4995
4948
6494
7667
8966
Imports
1060
841
1142
1316
1161
1046
Exports
7552
8062
8634
9555
10389
11664
Imports
8768
10267
9994
9538
7880
9917
Exports
10622
10575
12656
10116
8754
14124
Advertising
Computer and Information Services
Research, Development and Testing Services
Management Consulting and Public Relations
Services
Imports
Exports
Legal Services
Construction, Architectural and Engineering
Services
Imports
Exports
Industrial Engineering
Installation, Maintenance and Repair of Equipment
Operational Leasing
Other (1)
(1) Other includes accounting, auditing, and bookkeeping services; medical services; mining services; sports, and performing
arts; trade-related services; training services.
Source: Bureau of Economic Analysis, U.S. International Services Cross-Border Trade
Accessed at: http://www.bea.gov/international/intlserv.htm
101
Figure 4
102
Table A.1 Top 10 Occupations (by employment), by Industry, 2003 and 2008
Industry
NAICS Occupation
SOC
(1)
Legal Services
Accounting &
Bookkeeping Services
541100 All
Within NAICS
employment change,
2003-2008
(2)
Change in share of
NAICS employment
National-level
employment change
(3)
(4)
Movability index
ranking
(5)
00-0000
0.032
Lawyers
23-1011
0.046
0.004
0.073
180
Paralegals and legal
assistants
23-2011
0.226
0.025
0.224
41
Title examiners, abstractors, 23-2093
and searchers
0.580
0.008
0.241
21
Bookkeeping, accounting,
and auditing clerks
43-3031
-0.015
-0.001
0.060
28
File clerks
43-4071
0.099
0.001
-0.179
186
Receptionists and
information clerks
43-4171
0.077
0.001
0.037
208
Executive secretaries and
administrative assistants
43-6011
-0.264
-0.007
0.051
293
Legal Secretaries
43-6012
-0.023
-0.011
-0.024
148
Secretaries, except legal,
medical, and executive
43-6014
-0.290
-0.017
0.014
273
Office clerks, general
43-9061
0.031
0.000
-0.007
209
00-0000
0.113
541200 All
0.060
0.060
541200 Accountants and auditors
13-2011
0.279
0.042
0.226
40
541200 Tax preparers
13-2082
0.279
0.009
0.250
70
541200 First-line
supervisors/managers of
office and administrative
support workers
43-1011
-0.063
-0.004
-0.006
556
541200 Bill and account collectors
43-3011
1.114
0.010
-0.020
118
541200 Billing and posting clerks
and machine operators
43-3021
0.486
0.011
0.051
45
541200 Bookkeeping, accounting,
and auditing clerks
43-3031
0.154
0.004
0.060
28
103
Industry
NAICS Occupation
SOC
(1)
Architectural &
Engineering Services
Specialized Design
Services
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
541200 Payroll and timekeeping
clerks
43-3051
2.208
0.018
0.046
34
541200 Executive secretaries and
43-6011
0.164
0.001
0.051
293
541200 Secretaries, except legal,
43-6014
-0.001
-0.004
0.014
273
-0.004
-0.007
209
541200 Office clerks, general
43-9061
0.029
541300 All
00-0000
0.172
541300 Engineering managers
11-9041
0.100
-0.002
-0.065
228
541300 Architects, except landscape 17-1011
and naval
0.267
0.005
0.220
258
541300 Surveyors
17-1022
0.091
-0.002
0.083
261
541300 Civil engineers
17-2051
0.356
0.013
0.267
363
0.060
541300 Mechanical engineers
17-2141
0.371
0.005
0.124
93
541300 Architectural and civil
drafters
17-3011
0.276
0.005
0.175
104
541300 Civil engineering
technicians
17-3022
0.124
-0.001
-0.021
55
541300 Surveying and mapping
technicians
17-3031
0.337
0.004
0.246
67
43-6011
0.350
0.004
0.051
293
43-9061
0.208
0.001
-0.007
209
541400 All
00-0000
0.195
.
0.060
.
541400 General and operations
managers
11-1021
-0.081
-0.007
-0.103
650
541400 Art directors
27-1011
0.178
0.000
0.403
131
541400 Commercial and industrial
designers
27-1021
0.394
0.003
-0.013
125
104
Industry
NAICS Occupation
SOC
(1)
Computer Systems
Design & Related
Services
Within NAICS
employment change,
2003-2008
(2)
Change in share of
NAICS employment
(3)
National-level
employment change
(4)
Movability index
ranking
(5)
541400 Graphic designers
27-1024
0.273
0.013
0.377
103
541400 Interior designers
27-1025
0.541
0.034
0.152
326
541400 Sales representatives,
services, all other
41-3099
0.507
0.006
0.617
150
and auditing clerks
43-3031
0.077
-0.003
0.060
28
43-6011
0.251
0.001
0.051
293
medical and executive
43-6014
0.384
0.003
0.014
273
43-9061
0.216
0.001
-0.007
209
541500 All
00-0000
0.284
0.060
.
541500 Computer and information
systems managers
11-3021
0.355
0.002
0.041
301
541500 Management analysts
13-1111
0.404
0.002
0.264
223
541500 Computer programmers
15-1021
0.268
-0.001
-0.087
59
541500 Computer software
15-1031
0.325
0.004
0.260
14
15-1032
engineers, systems software
0.321
0.002
0.336
64
541500 Computer support
specialists
15-1041
0.212
-0.004
0.129
157
541500 Computer systems analysts
15-1051
0.462
0.011
0.032
9
541500 Network and computer
systems administrators
15-1071
0.280
0.000
0.378
87
541500 Network systems and data
communications analysts
15-1081
0.461
0.003
0.557
348
541500 Customer service
representatives
43-4051
-0.016
-0.007
0.174
161
105
Industry
NAICS Occupation
SOC
(1)
Management &
Technical Consulting
Services
Office administrative
services
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
541600 All
00-0000
0.292
.
0.060
.
managers
11-1021
0.119
-0.005
-0.103
650
13-1111
0.534
-0.001
0.264
223
541600 Business operations
specialists, all other
13-1199
0.897
0.011
0.216
225
541600 Market research analysts
19-3021
0.791
0.008
0.618
11
services, all other
41-3099
0.929
0.008
0.617
150
and auditing clerks
43-3031
0.184
-0.003
0.060
28
representatives
43-4051
0.041
-0.001
0.174
161
admin. Assistants
43-6011
0.334
-0.001
0.051
293
43-6014
0.281
-0.006
0.014
273
43-9061
0.142
-0.015
-0.007
209
561100 All
00-0000
0.338
0.000
.
.
managers
11-1021
0.446
0.004
-0.103
650
13-1111
0.195
-0.004
0.264
223
561100 Accountants and auditors
13-2011
0.855
0.011
0.226
40
561100 First-line
support workers
43-1011
0.551
0.005
-0.006
556
561100 Billing and posting clerks
and machine operators
43-3021
0.529
0.003
0.051
37
106
Industry
NAICS Occupation
SOC
(1)
Employment Services
Business Support
Services
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
and auditing clerks
43-3031
0.652
0.010
0.060
28
representatives
43-4051
0.950
0.022
0.174
161
43-6011
0.350
0.000
0.051
293
43-6014
0.753
0.007
0.014
273
209
43-9061
0.317
-0.001
-0.007
561300 All
00-0000
0.033
0.000
0.060
561300 Employment, recruitment,
and placement specialists
13-1071
1.110
0.013
0.255
440
561300 Registered nurses
29-1111
0.307
0.006
0.132
741
representatives
43-4051
0.471
0.009
0.174
161
43-6011
0.109
0.001
0.051
293
43-9061
0.033
0.000
-0.007
209
561300 Construction laborers
47-2061
0.687
0.012
0.218
567
561300 Team assemblers
51-2092
1.613
0.031
-0.006
405
561300 Helpers-production workers 51-9198
0.531
0.012
0.104
249
561300 Laborers and freight, stock,
and material movers, hand
53-7062
-0.189
-0.031
0.035
575
561300 Packers and packagers, hand 53-7064
-0.048
-0.003
-0.138
585
561400 All
00-0000
0.113
0.000
0.060
managers
11-1021
-0.030
-0.003
-0.103
650
561400 Medical transcriptionists
31-9094
0.094
0.000
-0.119
23
561400 Telemarketers
41-9041
0.101
-0.002
-0.146
42
107
Industry
NAICS Occupation
SOC
(1)
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
43-1011
0.206
0.003
-0.006
556
561400 Switchboard operators,
43-2011
including answering service
-0.221
-0.010
-0.293
233
561400 Bill and account collectors
43-3011
0.316
0.020
-0.020
118
representatives
43-4051
0.517
0.055
0.174
161
561400 Mail clerks and mail
machine operators, except
postal service
43-9051
0.244
0.002
-0.099
213
43-9061
0.011
-0.003
-0.007
209
561400 Office machine operators,
except computer
43-9071
0.088
-0.001
-0.122
537
511100 All
00-0000
-0.092
0.000
0.060
.
managers
11-1021
-0.103
0.000
-0.103
650
561400 First-line
support workers
Newspaper, book &
directory publishers
Within NAICS
employment change,
2003-2008
511100 Graphic designers
27-1021
0.408
0.016
-0.013
125
511100 Reporters and
correspondents
27-3022
-0.036
0.003
-0.035
40
511100 Editors
27-3041
0.051
0.014
0.009
147
511100 Advertising sales agents
41-3011
0.137
0.017
0.143
251
wholesale and
manufacturing, except
technical and scientific
products
41-4012
-0.170
-0.002
0.051
231
representatives
43-4051
-0.143
-0.002
0.174
161
108
Industry
NAICS Occupation
SOC
(1)
Software Publishers
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
511100 Mail clerks and mail
machine operators, except
postal service
43-9051
0.239
0.006
-0.099
213
43-9061
-0.112
-0.001
-0.007
209
511100 Printing machine operators
51-5023
0.016
0.003
0.019
145
511200 All
00-0000
0.074
0.000
0.060
.
0.095
0.001
0.041
301
systems managers
Motion picture & video
industries
Within NAICS
employment change,
2003-2008
11-3021
15-1021
-0.015
-0.006
-0.087
59
15-1031
-0.067
-0.022
0.160
14
15-1032
0.343
0.020
0.336
64
specialists
15-1041
-0.100
-0.014
0.129
157
15-1051
0.342
0.009
0.032
9
15-1071
0.493
0.007
0.378
65
511200 Computer specialists, all
other
15-1099
.
0.026
0.470
40
wholesale and
manufacturing, technical
and scientific products
41-4011
0.386
0.009
0.064
149
representatives
43-4051
0.312
0.005
0.174
161
512100 All
00-0000
0.019
0.000
0.060
.
managers
11-1021
-0.161
-0.005
-0.103
650
512100 Multi-media artists and
animators
27-1014
0.574
0.009
-0.043
122
109
Industry
NAICS Occupation
SOC
(1)
Sound recording
industries
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
(2)
(3)
(4)
Movability index
ranking
(5)
512100 Actors
27-2011
-0.622
-0.051
-0.144
640
512100 Producers and directors
27-2012
1.005
0.032
0.436
415
512100 Film and video editors
27-4032
0.245
0.006
0.240
257
512100 Counter attendants,
cafeteria, food concession,
and coffee shop
35-3022
0.017
0.000
0.143
521
512100 Motion picture
projectionists
39-3021
0.165
0.003
-0.024
518
512100 Ushers, lobby attendants,
and ticket takers
39-3031
-0.066
-0.009
-0.025
451
512100 Cashiers
41-2011
0.001
-0.001
0.024
396
43-6011
-0.022
-0.001
0.051
293
512200 All
00-0000
-0.229
0.000
0.060
.
managers
11-1021
-0.081
0.006
-0.103
650
27-2012
0.622
0.019
0.436
415
512200 Audio and video equipment 27-4011
technicians
-0.017
0.006
0.210
420
512200 Sound engineering
technicians
27-4014
1.238
0.116
0.402
252
services, all other
41-3099
0.645
0.013
0.617
175
wholesale and
manufacturing, except
technical and scientific
products
41-4012
-0.273
-0.002
0.051
231
and auditing clerks
43-3031
1.000
0.028
0.060
28
43-6011
0.000
0.008
0.051
293
110
Industry
NAICS Occupation
SOC
(1)
Radio & Television
Broadcasting
Cable & Other
subscription
programming
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
medical and executive
43-6014
0.000
0.008
0.014
273
43-9061
0.000
0.012
-0.007
209
515100 All
00-0000
-0.016
0.000
0.060
.
managers
11-1021
-0.125
-0.003
-0.103
650
27-2012
0.285
0.021
0.436
415
515100 Radio and television
announcers
27-3011
-0.105
-0.014
-0.100
204
515100 Broadcast news analysts
27-3021
-0.014
0.000
-0.089
163
515100 Reporters and
correspondents
27-3022
-0.046
-0.001
-0.035
40
515100 Broadcast technicians
27-4012
0.066
0.008
0.024
214
515100 Camera operators,
television, video, and
motion picture
27-4031
-0.215
-0.008
-0.101
439
41-3011
0.023
0.005
0.143
251
43-6011
0.071
0.001
0.051
293
43-9061
0.151
0.004
-0.007
209
515200 All
00-0000
-0.074
0.000
0.060
.
27-2012
0.275
0.021
0.436
415
515200 Audio and video equipment 27-4011
technicians
0.860
0.010
0.210
420
515200 Broadcast technicians
27-4012
0.635
0.013
0.024
214
41-3011
0.429
0.008
0.143
251
111
Industry
NAICS Occupation
SOC
(1)
Wired
telecommunications
carriers
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
services, all other
41-3099
0.206
0.009
0.617
175
representatives
43-4051
-0.443
-0.060
0.174
161
43-6011
0.207
0.005
0.051
293
515200 First-line
mechanics, installers, and
repairers
49-1011
-0.126
-0.001
-0.004
760
515200 Telecommunications
equipment installers and
repairers, except line
installers
49-2022
0.546
0.032
-0.002
359
515200 Telecommunications line
installers and repairers
49-9052
-0.016
0.008
0.135
523
517100 All
00-0000
0.133
0.000
0.060
.
13-1199
0.259
0.002
0.216
225
system administrators
15-1071
0.981
0.010
0.378
65
15-1081
4.900
0.028
0.557
348
517100 Electronics engineers,
except computer
17-2072
-0.038
-0.004
0.019
15
services, all other
41-3099
0.471
0.014
0.617
175
517100 Telephone operators
43-2021
-0.331
-0.014
-0.496
108
representatives
43-4051
0.260
0.010
0.174
161
112
Industry
NAICS Occupation
SOC
(1)
Wireless
telecommunications
carriers
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
517100 First-line
mechanics, installers, and
repairers
49-1011
0.038
-0.002
-0.004
760
installers
49-2022
0.097
-0.005
-0.002
359
49-9052
1.118
0.062
0.135
523
517200 All
00-0000
0.029
0.000
0.060
.
13-1199
0.799
0.008
0.216
225
specialists
15-1041
0.700
0.007
0.129
157
15-1081
-0.049
-0.001
0.557
348
except computer
17-2072
0.340
0.006
0.019
15
517200 First-line
retail sales workers
41-1011
0.451
0.011
0.009
739
517200 Retail salespersons
41-2031
1.138
0.088
0.109
460
services, all other
41-3099
0.235
0.012
0.617
175
517200 First-line
support workers
43-1011
-0.224
-0.007
-0.006
556
representatives
43-4051
0.387
0.066
0.174
161
113
Industry
NAICS Occupation
SOC
(1)
Satellite
telecommunications
Other
telecommunications
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
installers
49-2022
-0.022
-0.002
-0.002
359
517400 All
00-0000
-0.259
0.000
0.060
.
13-1199
1.071
0.015
0.216
225
15-1081
9.750
0.033
0.557
348
except computer
17-2072
0.933
0.015
0.019
15
services, all other
41-3099
0.061
0.025
0.617
175
517400 Telemarketers
41-9041
0.144
0.029
-0.146
42
representatives
43-4051
-0.356
-0.016
0.174
161
517400 Order clerks
43-4151
-0.500
-0.019
-0.182
80
installers
49-2022
-0.663
-0.033
-0.002
359
517400 Electrical and electronics
repairers, commercial and
industrial equipment
49-2094
1.583
0.018
-0.078
325
49-9052
-0.229
0.002
0.135
523
517900 All
00-0000
18.831
0.000
0.060
.
13-1199
93.500
0.021
0.216
225
specialists
15-1041
80.250
0.017
0.129
157
114
Industry
NAICS Occupation
SOC
(1)
Data processing &
related services
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
15-1081
58.333
0.017
0.557
348
except computer
17-2072
9.364
-0.022
0.019
15
517900 Retail salespersons
41-2031
.
.
0.109
460
services, all other
41-3099
78.400
0.063
0.617
175
wholesale and
manufacturing, technical
and scientific products
41-4011
95.000
0.021
0.064
149
representatives
43-4051
70.529
0.062
0.174
161
installers
49-2022
51.974
0.091
-0.002
359
49-9052
25.500
0.013
0.135
523
518200 All
00-0000
-0.081
0.000
0.060
systems managers
11-3021
0.108
0.006
0.041
301
15-1021
-0.471
-0.028
-0.087
59
15-1031
0.084
0.007
0.160
14
15-1032
0.539
0.022
0.336
64
specialists
15-1041
-0.076
0.000
0.129
157
15-1051
0.008
0.006
0.032
9
15-1071
0.062
0.005
0.378
65
115
Industry
NAICS Occupation
SOC
(1)
Other information
services
Within NAICS
employment change,
2003-2008
Change in share of
NAICS employment
National-level
employment change
Movability index
ranking
(2)
(3)
(4)
(5)
representatives
43-4051
-0.168
-0.006
0.174
161
518200 Computer operators
43-9011
-0.088
0.000
-0.329
170
518200 Data entry keyers
43-9021
-0.167
-0.007
-0.195
19
519100 All
00-0000
1.730
0.900
0.060
15-1021
16.714
0.028
-0.087
59
15-1031
18.227
0.031
0.160
14
15-1032
21.290
0.051
0.336
64
15-1081
16.381
0.027
0.557
348
519100 Librarians
25-4021
0.178
0.041
-0.014
453
519100 Library technicians
25-4031
-0.029
0.031
0.042
242
519100 Editors
27-3041
2.712
0.036
0.009
147
services, all other
41-3099
10.561
0.048
0.617
175
representatives
43-4051
1.958
0.039
0.174
161
519100 Library assistants, clerical
43-4121
0.660
0.067
0.044
352
116
References
Abraham, Katharine G. 1988. “The Role of Flexible Staffing Arrangements in Employers’ ShortTerm Adjustment Strategies.” In Employment, Unemployment and Hours of Work, Robert
A. Hart, ed. London: George Allen & Unwin, pp. 288–311.
———. 1990. “Restructuring the Employment Relationship: The Growth of Market-Mediated
Work Arrangements.” In New Developments in the Labor Market: Toward a New
Institutional Paradigm, Katharine G. Abraham and Robert B. McKersie, eds. Cambridge,
MA: MIT Press, pp. 85–119.
Abraham, Katharine G., and James R. Spletzer. 2009. “Addressing the Demand for Time Series
and Longitudinal Data on Occupational Employment.” Paper prepared for the conference
“Measurement Issues Arising from the Growth of Globalization,” held in Washington, DC,
November 6.
Autor, David, Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of Recent
Technological Change: An Empirical Exploration.” Quarterly Journal of Economics
118(4): 1279–1334.
Bardhan, Ashok Deo, and Cynthia A. Kroll. 2003. “The New Wave of Outsourcing.” Report
Series No. 1103. Fisher Center for Real Estate and Urban Economics. Berkeley, CA:
University of California.
Blinder, Alan S. 2006. “Offshoring: The Next Industrial Revolution?” Foreign Affairs 85(2):
113–128.
Blinder, Alan S. 2007. “How Many U.S. Jobs Might Be Offshorable?” CEPS Working Paper No.
142. Princeton, NJ: Princeton University.
Bureau of Labor Statistics (BLS). 2006. “Occupational Projections and Training Data, 2006–07
Edition.” Bulletin 2602. Washington, DC: BLS, February.
———. 2009. “National Employment and Wage Data for the 10 Largest Occupations by
Industry from the Occupational Employment Statistics Survey, May 2008.” Washington,
DC: BLS. http://stats.bls.gov/oes/ (October 26, 2009).
Dey, Matthew, Susan Houseman, and Anne Polivka. Forthcoming. “What Do We Know about
Contracting Out in the United States? Evidence from Household and Establishment
Surveys.” In Labor in the New Economy, Katharine G. Abraham, James R. Spletzer, and
Michael Harper, eds. Chicago: University of Chicago Press.
Dossani, Rafiq, and Martin Kenney. 2003. “Went for Cost, Stayed for Quality? Moving the Back
Office to India.” Working paper. Stanford, CA: Walter H. Shorenstein Asia-Pacific
Research Center, Stanford University, November.
117
Houseman, Susan. 2001. “Why Employers Use Flexible Staffing Arrangements: Evidence from
an Employer Survey.” Industrial and Labor Relations Review 55(1): 149–170.
Jensen, J. Bradford and Lori G. Kletzer. 2006. “Tradable Services: Understanding the Scope and
Impact of Services Offshoring.” In Brookings Trade Forum 2005, Offshoring WhiteCollar Work, Susan M. Collins, and Lael Brainard, eds. Brookings Institution:
Washington, DC, pp. 75–134.
———. Forthcoming.. “Measuring Tradable Services and the Task Content of Offshorable
Services Jobs.” In Labor in the New Economy, Katharine G. Abraham, James R. Spletzer,
and Michael Harper, eds. Chicago: University of Chicago Press.
Moncarz, Roger J., Michael G. Wolf, and Benjamin Wright. 2008. “Service-Providing
Occupations, Offshoring, and the Labor Market.” Monthly Labor Review 131(12): 71–86.
National Academy of Public Administration (NAPA). 2006a. Off-Shoring: An Elusive
Phenomenon. Washington, DC: National Academy of Public Administration, January.
———. 2006b. Off-Shoring: How Big Is It? Washington, DC: National Academy of Public
Administration, October.
———. 2007. Off-Shoring: What Are Its Effects? Washington, DC: National Academy of Public
Administration.
National Center for O*Net Development. 2006. “The O*Net Content Model.”
http://www.onetcenter.org/content.html (accessed June 8, 2010).
Peterson, Norman G., and Michael D. Mumford, et al. 1999. An Occupational Information
System for the 21st century: The Development of O*Net. Washington, DC: American
Psychological Association.
Peterson, Norman G., and Michael D. Mumford et al. 2001. “Understanding Work Using the
Occupational Information Network (O*Net): Implications for Practice and Research.”
Personnel Psychology 54: 451–492.
Segal, Lewis M., and Daniel G. Sullivan. 1997. “The Growth of Temporary Service Work.”
Journal of Economic Perspectives 11(2): 117–136.
van Welsum, Desiree, and Xavier Reif. 2006. “Potential Offshoring: Evidence from Selected
OECD Countries.” In Brookings Trade Forum 2005, Offshoring White-Collar Work,
Susan M. Collins and Lael Brainard, eds. Washington, DC: Brookings Institution, pp.
165–194.
Yuskavage, Robert E., Erich H. Strassner, and Gabriel W. Medeiros. 2009. “Outsourcing and
Imported Services in BEA’s Industry Accounts.” In International Trade in Services and
Intangibles in the Era of Globalization, Marshall Reinsdorf and Matthew Slaughter, eds.
Chicago: University of Chicago Press, pp. 247–288.
118
SESSION 3: MEASUREMENT IMPLICATIONS OF
TRANSFORMATION OFFSHORING AND
IMPORTED INTERMEDIATE INPUTS
Chair: Ned Howenstein (Bureau of Economic Analysis)
Outsourcing, Offshoring, and Trade: Identifying Foreign Activity
across Census Data Products,
Ron Jarmin, C. Krizan, and John Tang (U.S. Census Bureau)
Effects of Imported Intermediate Inputs on Productivity.
Lucy P. Eldridge and Michael J. Harper (Bureau of Labor Statistics)
Discussant: John Haltiwanger
119
Outsourcing, Offshoring, and Trade: Identifying Foreign
Activity across Census Data Products
Ron Jarmin, C.J. Krizan, and John Tang*
U.S. Census Bureau
May 2010
Abstract
The 2007 Economic Census asked establishments to identify if they engaged in domestic
outsourcing or foreign offshoring for manufacturing and wholesaling. These novel data can be
linked to existing longitudinal business microdata that include information on such variables as
employment, firm structure, and revenue. In this paper, we describe the collected responses,
their distribution across sectors, and some business activity patterns with reference to the U.S.
economy as a whole. We find that the majority of establishments do not offshore but those that
do are likely to belong to larger firms; furthermore, most offshorers can be linked to at least one
import transaction. Interestingly, less than a third of manufacturing activity occurs among
“traditional manufacturers”—firms that design and produce their own good and whose primary
activity is the production of their own goods. We observe additional differences in employment
shares and growth between offshorers and own producers. Finally, we find the special inquiry
data are a valuable complement to other Census Bureau microdata on trade transactions and firm
dynamics. While there is still more work needed to develop a fully integrated data infrastructure,
this paper demonstrates that analytic utility of that infrastructure will likely be very high.
*
Center for Economic Studies, U.S. Census Bureau, 4600 Silver Hill Road, Washington, DC 20233. Any opinions
and conclusions expressed herein are those of the authors and do not necessarily represent the views of the U.S.
Census Bureau. All results have been reviewed to ensure that no confidential information is disclosed. We would
like to thank Susan Houseman, Ken Ryder, Dennis Shoemaker, John Murphy, and participants in the National
Academy of Public Administration preconference workshop on measurement and globalization.
121
INTRODUCTION
Generally speaking, the practice by which firms transfer all or part their production to another
company is called “outsourcing” if the partner business is domestic and “offshoring” if
foreign.49 While offshoring and outsourcing have been controversial topics in the public
discourse, some have noted that these practices can impact many of the key measures we use to
track the health of our economy. Houseman (2007, 2008) notes in particular that increased
sourcing of imports, whose prices are poorly measured and biased upwards, is leading to
mismeasurement of industry productivity statistics. In addition, the growing ease with which
production activities can be moved around the globe to take advantage of factor price
differentials has made the classification and measurement of activity at domestic business
establishments and firms more difficult.
Our ability to quantify and examine how outsourcing and offshoring affect our statistics and
economy has been severely limited by a lack of appropriate data.50 This paper is an exploratory
study that utilizes a unique new dataset linking survey based offshoring data from the 2007
Economic Census with administrative import and export transactions files. With these data, we
are able to conduct a number of exercises aimed at assessing our ability to identify firms
engaging in these practices and to appropriately classify their activities in official statistics.
Moreover, the data we use are part of a broader effort under way at the Center for Economic
Studies (CES) to link import, export, outsourcing, and longitudinal firm data. The CES
maintains and updates an innovative dataset of the universe of transaction-level foreign trade
data linked to firm-level data from the longitudinal business database (LBD), the economic
censuses and other data sources. The new file is called the longitudinal firm trade transactions
database or (LFTTD). 51
In this paper, we describe and evaluate the new census 2007 questions on outsourcing. For
example, we break down the responses by industry sector and firm size and identify some
intuitively appealing stylized facts. We observe, for example, that although most offshoring
firms are small, offshoring firms are overrepresented among the largest firms, i.e., those that
employ more than 500 workers. To gauge the reliability of the offshoring responses, we match
them to international trade data and find that, as expected, a disproportionate share of the firms
that report outsourcing activities also can be linked to an import transaction. Finally, we more
closely scrutinize differences in employment shares and growth among the three different types
of manufacturing sourcing firms.
The paper proceeds as follows: the next two sections describe the new data we use and provide
some basic statistics and various exercises that include data quality checks; disaggregation by
major sector, size, and activity; linkage to trade data; comparisons at the establishment- and firm49
Technically, this usage is imprecise as outsourcing can refer to both domestic and foreign, so offshoring is more
accurately defined as foreign outsourcing. For ease of reference, this paper will follow the less precise, conventional
usage.
50
Helpman (2006) notes that while theoretical work on why firms outsource production or invest abroad for vertical
integration is inconclusive, very few studies empirically test some model implications due to a lack of appropriate
data.
51
The LFFTD was developed primarily through the efforts of J. Bradford Jensen, an RDC researcher. See the data
appendix in Bernard, Jensen, and Schott (2009) for details.
122
levels; and preliminary analyses of employment differences. The last section summarizes our
findings.
DATA
Among other business microdata, the CES maintains and updates a novel dataset of the universe
of transaction-level foreign trade data linked to firm-level data from the LBD, the business
register, economic censuses, and other data sources. The transaction-level international trade
files, also known as the foreign merchandise trade (FMT) data, underlie the census bureau’s
published foreign trade statistics, which are the official source of data on U.S. international
trade.52 The export data come from exporters’ electronic filings on the census bureau’s
automated export system and also through a data-sharing arrangement with the Canadian
government. Each filing represents a shipment of one or more kinds of merchandise from one
exporter to one importer on a single carrier. Similarly, the import data come from the U.S.
Customs’ Automated Commercial System, which collects information on imports from import
entry forms, warehouse withdrawal forms, and foreign trade zone documents.
These data contain information for each transaction, including the 10-digit harmonized schedule
code, value, quantity, entry or exit port, date of transaction, mode of transportation, and relatedparty status. Data are collected for every import transaction with a value greater than $2,000 and
every export transaction with a value greater than $2,500. In addition, the employer
identification number of the importer or exporter is collected. This is the primary variable used
to link the records to other census data products like the business register.53
The business register is the primary file used to assign firm identifications to transaction-level
trade data. In particular, it contains establishment-level data including employer identification
number, firm name, firm identification, address and industry affiliation. Matching transactionlevel import data to firm identifications is relatively straightforward. Because most export and
all import transaction data contain a field for the employer identification number, observations
can be linked directly to the business register. The match rates of import transactions to the
business registerare typically in the 80 percent range and the share of matched import value is
typically above 80 percent. The linked trade transaction data with firm identifiers are the key
components of the LFTTD.54
For this exercise, we link the LFFTD files to the special inquiries data on offshoring and
outsourcing in the 2007 economic census. The questions were originally designed to help census
more accurately describe firms’ supply chains and to aide in the classification of the increasingly
52
Tang (2009) describes the FMT in detail and provides useful information including variable definitions,
codebooks, and variable coverage over time. The data cleaning performed to construct the FMT include, for
example, assigning time-consistent variable names.
53
The EIN variable is not present on records of exports to Canada due to a bilateral data exchange program; instead,
name and address are used. Because of differences in matching methodologies as well as the sheer number of
records (20 million per year), it has taken several years for researchers to develop matching algorithms that can be
rapidly and reliably applied to new years of data.
54
The description of the matching procedure for imports and exports draws heavily on the data appendix in Bernard,
Jensen, and Schott (2009).
123
complex web of manufacturing activities. In particular, they ask manufacturing and wholesaling
plants whether they designed the goods they sell, their primary activity was manufacturing (for
themselves or others) or re-sales, and if they purchased contract-manufacturing services from
either foreign or domestic companies.55 For all establishments that received a form, about 72
percent of wholesale establishments and 66 percent of manufacturing plants responded to the
questions, which is roughly comparable to the 72 percent and 73 percent response rates for
employment. 56 Although the questions were not officially pretested, our results indicate that the
responses make intuitive sense.
EXPLORATORY EXERCISES
Special Inquiry Cross-Tabulations, by Activity, Count, and Size
We begin our analysis with a few basic tabulations. Table 1 presents establishment-level
breakdowns for each part of the special inquiry, while Table 2 contains analogous figures for
employment.57 From these basic summary statistics, a number of interesting patterns quickly
emerge. Almost 60 percent of establishments responding to this part of the manufacturing form
indicate that they design their own goods (row 2, column 1), while only 15 percent of responding
wholesale establishments state the same. For part 2 of the special inquiry, the majority of
responses to both forms are consistent with the expected industry definition of the
establishments. For example, roughly 81 percent of the manufacturing establishments (and about
86 percent of their employment) is accounted for by establishments that reported their major
activity as either “production” or “contracting.” Similarly, about 68 percent of establishments
(and 66 percent by employment) that answered the wholesale forms reported their major activity
as “resales.”58 Note that although the first row indicates the total number of establishments or
their associated employment, not all establishments necessarily answered each part of the
inquiry, which is why the subtotals for each part may sum to less than the total responses.
Interestingly, 5 percent of tabbed manufacturing establishments report that their primary activity
is resales and 7 percent of tabbed wholesale establishments report their major activity is
production. While the special inquiry data were not used for classification purposes, the
Economic Census is the most reliable source of industry codes available to the Census Bureau
and quinquennial collection results in a substantial number of corrections to establishment
55
2007 Economic Census, Forms MC-313XX through 315XX, Question 26. Although the question was asked in
the Census of Manufacturers, respondents included both manufacturers and wholesalers. See Appendix A at the end
of this paper for the specific questions (under “Special Inquiries”). Note that the language used in these questions
was not pretested.
56
These include all long-form manufacturing cases and all wholesale establishments except for Miscellaneous
wholesale and Agents and brokers. Furthermore, note that establishments receiving forms represent less than half of
the universe of establishments. We thank Dennis Shoemaker in the Census Bureau’s Economic Planning and
Coordination Division for providing the background information.
57
Since this categorization is based on which forms the establishments responded to and not on a comprehensive
measurement of the overall firms’ activities, one should interpret the table accordingly. That is, it may be that the
establishment is classified as “wholesale” but report doing their own production because they are part of a larger
firm that has manufacturing activities.
58
We should note that this sample conditions on the establishments answering all three questions and also excludes
wholesale establishments that are known to act primarily as manufacturers’ sales offices.
124
industry codes. In fact, about 5 percent of the manufacturing establishments (nonbank) were
classified in different (nonmanufacturing) sectors after the 2007 Economic Census form was
received and the share of wholesalers that switched sectors was roughly twice that. These
numbers are typical during Economic Census operations. Thus, the findings from the special
inquiries are in line with typical reclassification rates.
In light of recent research on the effects of professional employer organizations on industry
statistics, it is worthwhile to note that in separate calculations, we found these shares to have
been relatively stable since at least 1997.59 That is, it seems unlikely that the share of
manufacturers reporting that they engage in other activities is due to them reclassifying
themselves because they use professional employer organizations. Instead, it seems like a
normal part of the classification process. Furthermore, most establishments indicated that they
neither outsource nor offshore any activity. For manufacturing, the offshorer, outsourcer, and
own-producer shares were roughly 2, 26, and 69 percent, respectively. For wholesale
establishments, the analogous shares were 4, 11, and 82 percent.
The employment breakdowns in Table 2 are qualitatively similar to the figures in Table 1, with
employees concentrated in each sector’s primary functions (manufacturing and reselling).
However, note that establishments that offshore production—bidparticularly manufacturers—
represent twice their share of manufacturing employment (4 percent) than they do of
establishments (2 percent). Thus, establishments reporting offshoring activity are larger, on
average, than nonoffshorers.60
The decision to outsource or offshore production activities is better thought of as a firm-level
rather than establishment-level choice. For the exercises that follow, we will shift our unit of
analysis to the firm-level. In order to do this we need a protocol for aggregating the
establishment-based questions from the Economic Census. Our approach is to classify a firm as
an offshorer if it operates at least one establishment that reports offshoring activity, and as an
outsourcer if it has no offshoring establishments and at least one outsourcing establishment.
Firms with no contracts make up the balance.
One important firm characteristic that is likely related to the propensity to engage in outsourcing
and offshoring activity is size. Table 3 shows the number of firms grouped by size and primary
sourcing activity. While small firms (those with 50 or fewer employees) dominate each category
of firm, a greater proportion of offshoring firms (8 percent) employ more than 500 employees
compared to firms with no contracts (1 percent) or those using domestic suppliers (5 percent).
That is, while most offshoring firms are small, the greatest share of offshoring activity can be
found among large firms.
59
Dey, Houseman, and Polivka (2009).
The relative share of domestically outsourced employment in manufacturing (35.0 percent) reported here is
comparable to that found using the 2005 Contingent Worker Supplement (CWS) collected by the Bureau of Labor
Statistics (38.7 percent) (Dey, Houseman, and Polivka (2009, Table 3).
60
125
Matching to the Trade Transactions Files
A natural quality control check on the validity of the offshoring responses was to look at
differences in observed importing activities among the three firm production types. As shown in
Table 4, the overwhelming majority (78 percent) of firms that reported offshoring activity on the
2007 Economic Census form can be matched to at least one import transaction in 2007. We
cannot conclude that most of the remaining firms that responded that they are offshorers but
were not matched to an import transaction answered the form incorrectly because 1) our
matching methodology between firms and import transactions is still under development, and 2)
not all firms that outsource their production will necessarily reimport the good. Clearly many of
these firms are multinational corporations that sell goods and services in many different
countries, and it may be that the majority of their offshore production is aimed at foreign
markets. Finally, 3) many firms use third-party wholesale firms to handle their foreign trade
activity.
Interestingly, Table 4 also shows that while the shares of total import value are fairly similar
across the production categories, if one considers the much smaller number of offshoring firms it
is clear that offshorers import far more than other types of producers. We also corroborated this
hierarchy of import activity by firm production type by focusing on two specific industries:
vehicle manufacturing and electronics, where we found even stronger results.61 Additional
breakdowns by each census form type are presented in Tables 5 and 6, which also present the
number of firms and their employment associated with the responding establishments. We in
turn identify the number of manufacturing and wholesaling establishments owned by these firms,
which may or may not have responded to the census forms, as well as how many of these firms
can be identified in the import transaction data and what share of total U.S. imports those
transactions represent. It is interesting to note the employment discrepancy among respondents
to the wholesaling forms (Table 6), with both own producers and outsourcers having a larger
share of their workers in wholesaling establishments (column 4) than in manufacturing, as one
would expect; yet among offshorers this is not the case. These statistics hint at the complex
structure of larger multinational companies. We’re hopeful that the integrated data infrastructure
discussed here and still under development will help researchers to get a better handle on how
firms and value chain evolve and the role outsourcing and offshoring play in this evolution.
Comparing Manufacturing Types
In a recent federal register notice the Organization of Management and Budget’s (OMB’s)
Economic Classification Policy Committee (ECPC) pointed out some of the difficulties involved
in defining what a manufacturing establishment is in the presence of outsourcing and offshoring
(OMB 2009). They define three general types of manufacturing units: 1) traditional or integrated
manufacturing, 2) manufacturing service providers, and 3) factoryless goods providers. They
define the major characteristics of each as follows:
61
Results are withheld due to potential disclosure of confidential information.
126
1) Traditional manufacturers
•
•
•
•
2)
Manufacturing service providers
•
•
•
•
•
•
3)
Perform transformation activities
Own the rights to the product they manufacture
Control and facilitate the production process
Sell the final good
Performs transformation activities
Receives contracts to perform transformation activities
Does not own intellectual property or design of the final product
Does not own the final product
Controls the production facility but not the production process
Does not sell the final product
Factoryless goods providers
•
•
•
•
•
Does not perform transformation activities
Contracts with manufacturing service providers
Owns the intellectual rights to the final product
Owns the final product
Sells the final product
We make use of the new questions on the 2007 Economic Census to approximate these
categories using the following definitions:
1)
Traditional manufacturers
•
•
•
2)
Establishment does not contract out for manufacturing services from other
companies or other establishments of its company
Establishment’s primary activity is manufacturing
Establishment designs, engineers, or formulates the manufactured products it sold,
produced or shipped
•
•
•
Establishment does not contract out for manufacturing services from other
companies or other establishments of it’s company
Establishment’s primary activity is providing contract services for others
Establishment does not design, engineer, or formulate the manufactured products
it produced or shipped
127
3)
Factoryless goods provider
•
•
•
Establishment contracts out for manufacturing services from other companies or
other establishments of its company (both in and outside of the United States
Establishment’s primary activity is resales
Establishment designs, engineers, or formulates the manufactured products it sold,
produced or shipped
Table 7 displays the shares of activity accounted for by each establishment type. The
denominator is the sample of total activity accounted for by the establishments that answered all
three questions. While traditional manufacturers dominate the three categories of producer types,
they represent less than a third of total manufacturing activity by establishment count and
employment.
Of course, it may be that traditional manufacturers are more (or less) common in certain
industries. We investigate this possibility by calculating the employment shares accounted for
by these producers for a subset of NAICS industries. Due to disclosure concerns, we limit the
analysis to only those industries with a relatively large number of firms; results are displayed in
Figure 1.62
Clearly, there is wide variability in the shares of activity accounted for by traditional
manufacturers across industry subgroups. The range of activity starts at about 8 percent for
printing and ranges to almost half for textile mills. Interestingly, computer manufacturing, an
industry one would normally associate with outsourcing, is only slightly above the average share
of employment at traditional manufacturers. This unexpected finding for computer
manufacturing suggests that as outsourcing and offshoring become more common, firms may
become less manufacturing intensive over time. That is, firms that previously had a large share
of manufacturing employment may begin to specialize more heavily in other activities and it may
affect their manufacturing employment, overall employment or both.
To explore this issue further, we identified firms that existed in both 1990 and 2007 and
categorized them according to whether or not they outsourced, offshored, or did not contract out
for manufacturing services. Next we calculated the firms’ shares of manufacturing employment
in both years as well as the changes in total employment for each group; these are shown in
Figures 2 and 3, respectively.
Both figures show an ordering to the changes in employment. The firms without contracts
decline less or grow more than either outsourcers or offshorers. In Figure 2 noncontracting
firms’ manufacturing shares (weighted means) declined about 13 percent, similar to outsourcers
(14 percent) but visibly less than offshorers (18 percent). Similarly, in Figure 3 we see that own
producers had much stronger growth than did either outsourcing or offshoring firms. In fact,
employment actually declined at offshoring firms.
62
We find similar results for establishment shares but omit them here for brevity.
128
Multivariate Analysis of Employment Trends
The above findings do not control for any of the many other factors that are known to affect firm
growth rates, such as age, size, and geography. While a rigorous treatment of these factors is
beyond the scope of this exploratory paper, we begin by running linear regressions of changes in
manufacturing shares and total employment on a set of basic firm controls, as well as firm type.
The results are reported in Table 8.
In both specifications, the omitted firm type is non-contractors so the results should be
interpreted accordingly. Interestingly, the regression results do not completely support our
preliminary observations from the earlier figures. Controlling for other major factors, both
outsourcers and offshorers are associated with a more negative change in their shares of
manufacturing employment (column 1, rows 5 and 6) relative to own producers. On the other
hand, with growth in total manufacturing employment as the dependent variable (column 2), it
appears that outsourcers may have had more employment growth than own producers, and
substantially more than offshorers. These discrepant findings suggest that substantial care must
be taken in any interpretation and that much more work is necessary before we are fully
confident in the results.
CONCLUSION
This paper takes advantage of a unique new dataset linking offshoring data from the 2007
Economic Census with import and export transactions files to examine the prevalence of
outsourcing and offshoring and how these activities are correlated with firm productivity. We
performed a number of preliminary quality control and exploratory exercises and obtained the
following six results:
1) The majority of establishments do not report either outsourcing or offshoring activity.
2) Most establishments’ activity is consistent with their industry definitions. That is, most
wholesalers report resales as their primary activity and most manufacturers report either
manufacturing or contracting as their primary activity. Differences from these norms are
in-line with historical industry changes that normally occur during economic censuses.
3) The majority of offshoring firms are small but large firms are more likely to engage in
offshoring.
4) We are able to match 78 percent of the firms that reported engaging in offshoring activity
to at least one import transaction. This is encouraging given that there is some noise in
our linking variable and that a firm that offshores does not necessarily need to reimport
the good.
5) Less than a third of manufacturing activity occurs at traditional manufacturing plants that
design and produce their own goods and whose primary activity is manufacturing for
themselves.
129
a. A further 11 percent occurs at manufacturing service providers.
b. Less than 1 percent is accounted for by factoryless goods providers
c. There are substantial differences in these shares across industries.
6) As a group, noncontractors grew more and stayed more manufacturing intensive than
both outsourcers and offshorers, but when we controlled for key firm characteristics we
found that outsourcing firms grew more than noncontractors.
Additional areas for study that can usefully exploit these data include examining the role of
outsourcing and offshoring in firm and productivity dynamics. Combing the data infrastructure
described in this paper with Bureau of Economic Analysis data on trade in services and foreign
direct investment would greatly enhance the analytic capability of both permitting the analysis
of, for example, changes in the distribution of manufacturing and wholesaling activities across
establishments within domestic-only and multinational firms, and investment patterns by sector
and firm type.63 Discussions about bringing these rich data sources together are ongoing.
Table 1: Special Inquiry Response Breakdown, by Form Type: Establishment
a
Total establishment count
Own design
Yes
No
Primary function
Production
Contracting
Reselling
Other
Primary sourcing
No contracts
Domestic outsourcing
Foreign offshoring
*
Manufacturing
106,550
Wholesaling
153,147
63,017
41,266
22,554
127,942
57,371
28,725
5,326
10,725
10,061
4,169
104,900
20,627
74,030
28,173
2,269
126,100
16,762
5,397
Includes all establishments that answered any part to special inquiry.
63
Data on foreign direct investment are collected by the Bureau of Economic Analysis; see Mataloni (1995),
Quijano (1990), and.Bureau of Economic Analysis (2004, 2006).
130
Table 2: Importing, Offshoring, and Outsourcing
No contracts
Foreign offshoring
2007 EC
respondent firms
154,961
33,313
6,055
2007 import data
matches
40,827
10,250
4,750
% Total value of
imports (2007)
19%
16%
14%
Table 3: Special Inquiry Response Breakdown, by Form Type: Employment
Total employee count
Own design
Yes
No
Primary function
Production
Contracting
Reselling
Other
Primary sourcing
No contracts
Foreign offshoring
Manufacturing
9,215,356
Wholesaling
2,805,397
6,129,012
2,937,883
547,211
2,209,371
6,211,181
1,790,786
198,641
603,521
291,188
63,166
1,847,761
422,633
5,486,210
3,161,254
361,009
2,195,630
380,445
143,767
Table 4: Number of Firms by Employment Size
No contracts
Foreign offshoring
≤50
122,139
23,899
4,402
51-500
21,401
7,699
1,122
131
>500
2,007
1,551
506
Table 5: Special Inquiry Part 3 Breakdown: Primary Sourcing in Manufacturing Responses
Estab.
Firms
Affiliated
mfg. estab.
Affiliated
wholesale
estab.
Matched
to import
data
Import
value*
($ bil)
39,442
1,371,578
20,738
16,070,271
$855.7
Total
Count
Employees
104,472
9,008,473
75,677
113,084
19,575,864 10,472,662
No contracts
Count
Employees
74,030
5,486,210
52,629
7,602,322
65,520
3,872,270
13,306
479,717
11,932
5,135,543
$205.9
Outsourcing
Count
Employees
28,173
3,161,254
20,991
7,579,528
37,021
4,276,430
14,806
499,135
7,088
6,588,885
$356.3
Offshoring
Count
employees
2,269
361,009
2,057
4,394,014
10,543
2,323,962
11,330
392,726
1,718
4,345,843
$293.5
*Note: total U.S. import value in 2007 = $2,344.6 billion
132
Table 6: Special Inquiry Part 3 Breakdown: Primary Sourcing in Wholesaling Responses
Estab.
Firms
Affiliated
mfg. estab.
Affiliated
wholesale
estab.
Matched to
import data
Import
valuea
($ bil)
Total
Count
Employees
148,259
2,719,842
122,227
16,431,116
24,966
4,746,349
194,312
3,839,155
37,586
14,316,785
$987.3
No contracts
Count
Employees
126,100
2,195,630
104,087
7,527,531
6,631
779,450
148,036
1,604,389
29,910
5,775,482
$387.9
Outsourcing
Count
Employees
16,762
380,445
13,700
4,728,984
10,634
2,015,647
29,948
749,150
4,223
4,413,939
$298.5
Offshoring
Count
Employees
5,397
143,767
4,440
4,174,601
7,701
1,951,252
16,328
485,616
3,453
4,127,364
$300.9
*
Note: Total U.S. import value in 2007 = $2,344.6 billion
Table 7: Activity Shares of Manufacturing Types
Traditional manufacturers
Factoryless manufacturers
All other manufacturing types
Establishment share (%)
28.1
11.0
0.9
60.0
133
Employment share (%)
30.3
6.0
0.5
63.2
Table 8: OLS Regression Results
DV:
Firm age
# States
1990 Firm employees (1,000)
1990 Estab. count (1,000)
Outsourcer
Offshorer
Intercept
(1)
Δ share of mfg. employment
(2)
% Δ in total mfg.
employment
-0.013***
(0.000)
-0.009***
(0.001)
-0.000
(0.001)
0.065***
(0.022)
-0.023***
(0.004)
-0.049***
(0.011)
0.287***
(0.004)
-0.044***
(0.001)
0.018***
(0.001)
-0.011***
(0.001)
0.072*
(0.042)
0.048***
(0.008)
-0.075***
(0.022)
0.684***
(0.008)
34,667
0.055
34,667
0.110
Observations
Adjusted R-squared
Significance levels: *10 percent **5 percent ***1 percent
Standard errors in parentheses.
134
Fa
br
ic P
Tr ate rin
a n d tin
s M g
Pr por et a
im t E ls
ar q
y uip
M
e
Ap ta
p a ls
re
M Pa l
ac p
e
C hin r
M om er
is p y
ce u
ll te
Be ane rs
ve ou
ra s
Pl ges
C as
he tic
Te E mi s
xt lec cal
ile E s
Pr qu
od ip
uc
W ts
oo
N
on P F d
m e oo
e t t ro d
al le
M um
Fu ine
rn ral
in s
Le ture
Te a
xt th e
ile r
M
ills
Figure 1
50
45
40
35
30
25
20
15
10
5
0
Traditional Manufacturing Employment Shares
Figure 2
Changes in Manufacturing Shares
0%
No Contracts
Outsource
-5%
-10%
-15%
-20%
135
Offshore
Figure 3
% Change in Total Employment
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
-10.0%
No Contracts
Outsource
-20.0%
136
Offshore
Appendix A
137
References
Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. 2009. “Importers, Exporters and
Multinationals: A Portrait of Firms in the U.S. that Trade Goods.” In Producer
Dynamics: New Evidence from Micro Data, T. Dunne, J.B. Jensen, and M.J. Roberts,
eds. Chicago: University of Chicago Press.
Bureau of Economic Analysis. 2006. “Foreign Direct Investment in the United States: Final
Results from the 2002 Benchmark Survey.” Washington, DC: Bureau of Economic
Analysis. http://www.bea.gov/international/pdf/fdius_2006p/Introduction_06P.pdf
(accessed October 14, 2009).
———. 2004. “U.S. Direct Investment Abroad: Final Results from the 1999 Benchmark
Survey.” Washington, DC: U.S. Government Printing Office.
http://www.bea.gov/scb/pdf/internat/usinvest/meth/usdia99.pdf (accessed June 2, 2010).
Dey, Matthew, Susan Houseman, and Anne Polivka. 2009. “What Do We Know about
Contracting Out in the United States? Evidence from Household and Establishment
Surveys.” Upjohn Institute Staff Working Paper 09-157. Kalamazoo, MI: W.E. Upjohn
Institute for Employment Research.
Helpman, Elhanan. 2006. “Trade, FDI, and the Organization of Firms.” Journal of Economic
Literature 44(3): 589–630.
Houseman, Susan. 2007. “Outsourcing, Offshoring, and Productivity Measurement in U.S.
Manufacturing.” International Labour Review 146(1–2): 61–80.
———. 2008. “Outsourcing and Offshoring: Problems for Price and Productivity
Measurement.” Photocopy. W.E. Upjohn Institute for Employment Research,
Kalamazoo, MI, June.
Mataloni, Raymond J., Jr. 1995. “A Guide to BEA Statistics on U.S. Multinational Companies.”
Survey of Current Business 75(3): 38–55.
Office of Management and Budget (OMB). 2009. “2007 North American Industry Classification
System (NAICS) Updates for 2012; Notice.” Federal Register 74(4).
http://www.census.gov/eos/www/naics/federal_register_notices/notices/fr07ja09.pdf
(accessed June 2, 2010).
Quijano, Alicia M. 1990. “A Guide to BEA Statistics on Foreign Direct Investment in the United
States.” Survey of Current Business February: 29–37.
http://www.bea.gov/scb/pdf/internat/fdinvest/1990/0290iid.pdf (accessed June 2, 2010).
Tang, John. 2009. “Technical Note to the Foreign Merchandise Trade Data.” Photocopy. Center
for Economic Studies, U.S. Census Bureau, Washington, DC.
139
Effects of Imported Intermediate Inputs on Productivity*
Lucy P. Eldridge and Michael J. Harper
U.S. Bureau of Labor Statistics
January 15, 2010
* The authors would like to thank Erich Strassner and Robert Yuskavage of the BEA for
providing the import data necessary for this study. We also thank Steve Rosenthal and Randy
Kinoshita for helpful comments and assistance. All views expressed in this paper are those of the
authors and do not necessarily reflect the views or policies of the U.S. Bureau of Labor Statistics.
Authors can be contacted via e-mail at [email protected] and [email protected], or
by mail at U.S. Bureau of Labor Statistics, 2 Massachusetts Ave., NE, Rm. 2150, Washington,
DC 20212.
141
There is significant interest in determining the effects of offshoring on U.S. economic
performance. Offshoring, or offshore outsourcing, is the substitution of imported intermediate
inputs for domestic labor or domestically produced intermediate inputs. To assess the effects of
imported intermediate inputs on Bureau of Labor Statistics (BLS) productivity statistics, it is
essential to understand how imports enter into the measurement framework. The BLS Major
Sector Productivity program develops measures of labor productivity for broad sectors of the
economy: business, nonfarm business, manufacturing, and nonfinancial corporations. In addition,
this program develops annual indexes of multifactor productivity (MFP) for the private business
sector, the manufacturing sector, and for most manufacturing groups. This paper focuses on the
BLS productivity measures for the private business sector, as well as the manufacturing sector.
Productivity measures for these two sectors are constructed using different methodologies; the
private business sector productivity measures use a value-added output concept, while the
manufacturing sector measures use a sectoral output approach. This difference in methodology
influences the effects of imported intermediates on the measures of productivity.
In this paper, we develop a framework for estimating the effects of imported intermediate inputs
on U.S. major sector labor productivity. The production model used to calculate the BLS private
business sector MFP measures is expanded to treat imported intermediate inputs as an input,
rather than as a subtraction from output. The BLS framework for constructing manufacturing
MFP is decomposed in order to isolate imported intermediate inputs. For both sectors, we use the
Solow MFP equation to estimate the effects on labor productivity of substitution between
imported intermediate inputs and U.S. hours worked (Solow 1957). The data reveal that the
growth in imported intermediate inputs contributed 14 percent to the average annual growth of
labor productivity for the private business sector from 1997 to 2006, and contributed 23 percent
to the average annual growth in labor productivity in the manufacturing sector.64
The study also addresses the difficulties surrounding the deflation of the imported intermediate
inputs, since the coverage of import price indexes is sparse. We show that any mismeasurement
of import prices will impact measured productivity growth. However, the size of the effect will
depend upon the share of imports relative to aggregate output, which range from 8–12 percent
for the private business sector and 12–18 percent for the manufacturing sector.
DATA SOURCES
Output
Real output measures used by the BLS to construct major sector productivity statistics are
produced by the Bureau of Economic Analysis (BEA) of the U.S. Department of Commerce.
The most widely known measure of aggregate output for the U.S. economy is gross domestic
product (GDP). GDP is the sum of 1) personal consumption expenditures, 2) gross private
domestic investment, 3) government consumption expenditures and gross investment, 4) exports
of goods and services, less 5) imports of goods and services. The BEA constructs nominal output
64
We implement the analysis using BEA annual input-output tables, as well as data on imported intermediate inputs
provided by the BEA. See Yuskavage, Strassner, and Medeiros (2008).
143
for detailed components of GDP from various data sources, converts them to real measures, and
then aggregates them to calculate GDP.
As a fundamental part of the national accounts, the BEA also distinguishes three primary sectors
of GDP: business, household, and government (Young and Tice 1985). The business sector
accounts for the bulk of national output. The BEA calculates the measure of business sector
output by removing from GDP the gross product of general government, private households, and
nonprofit institutions.65
Ideally, BLS productivity statistics would measure productivity for the U.S. economy at the most
aggregate level of domestic output, GDP. However, the BLS must exclude several activities from
aggregate output in order to remove potential sources of bias specific to productivity
measurement. The real gross products of general government, of private households, and of
nonprofit institutions are estimated primarily using data on labor compensation. The trends in
such output measures will, by definition, move with measures of input data and will tend to
imply little or no labor productivity growth. Although these measures are the best available
estimates of nonmarket components of GDP, including them in measures of aggregate
productivity for the economy would bias labor productivity trends toward zero.
The BLS private business sector also excludes the gross product of owner-occupied housing and
the rental value of buildings and equipment owned and used by nonprofit institutions serving
individuals.66 These components are excluded because no adequate corresponding labor input
measures have been developed. To measure MFP, the BLS must further restrict output to the
U.S. private business sector, excluding the output of government enterprises. Estimates of the
appropriate weights for labor and capital in government enterprises are not made because
subsidies account for a substantial portion of capital income; therefore, there is no adequate
measure of government enterprise capital income in GDP. In 2006, the BLS measure of the U.S.
private business sector output accounted for approximately 76 percent of the value of GDP.67
In the manufacturing sector, BLS measures output for productivity statistics differently. Output
in the manufacturing sector is the deflated value of production shipped to purchasers outside of
the domestic industry, not just production for final users as is used for the major sector MFP
indexes. This is a sectoral output concept, defining output as gross output excluding intrasectoral
transactions (sales or transfers between establishments within the sector)—sales to final demand
plus the intermediate goods sent to other industries. The manufacturing MFP indexes are based
on sectoral output in an effort to avoid the problem of double-counting that occurs when one
establishment provides materials used by other establishments in the same industry.
65
The gross product of general government is the sum of government expenditures on compensation of general
government employees and the general government consumption of fixed capital, which measures the services of
general government fixed assets. Government expenditures on goods and services purchased from the private sector
are not excluded from private business sector output. The gross product of private households is the compensation of
paid employees of private households; the gross product of nonprofit institutions serving individuals is the
compensation paid to employees of these institutions.
66
This value is measured as the sum of consumption of fixed capital, indirect business taxes, and interest paid.
67
Data in this paper originate in the MFP program and coverage differs from BLS quarterly labor productivity
measures for business sector. MFP measures are available only on an annual basis and exclude government
enterprises from sectoral coverage.
144
Labor Input
Labor input for the U.S. private business sector is measured as total hours actually worked by all
persons multiplied by a labor composition index. The hours actually worked measure is based on
the sources and methods used to measure quarterly business sector labor productivity. The BLS
labor composition index estimates the effects that shifts in age, education, and gender have on labor
input growth and MFP growth.
Labor input is based on a jobs concept. The Current Employment Statistics program (CES) is the
primary source of data used to construct hours for the BLS productivity measures.68 The CES
average weekly hours paid data are adjusted to an hours-at-work concept using a ratio of hoursworked to hours-paid.69 Current Population Survey (CPS) data on average weekly hours of
nonproduction and supervisory workers are incorporated into the methodology to expand
coverage to all employees.70 To expand sectoral coverage, hours actually worked for employees
of farms, proprietors, and unpaid family workers reported in the CPS are incorporated into the
labor input measure; remaining data are obtained from various sources.71
Construction of the MFP labor composition measure begins with estimates of the number of
hours worked by each type of worker based on CPS data. The assembles data on workers’ hours
classified by their educational attainment, age, and gender using actual wage averages for
weights. The sum over all groups of the hour’s growth rates multiplied by the labor cost shares
gives the growth in adjusted labor input. Subtracting this from the growth in total (unweighted)
hours yields the growth in labor composition.72
Labor input for the U.S. manufacturing sector is constructed using the same methods, except that
no adjustment is made for labor composition (age, education, and gender of the work force). The
labor composition adjustment is currently not included in manufacturing hours data due to
limitations in the data available from the CPS at the industry detail.73
68
The CES, an establishment survey, sample is benchmarked annually to levels based on administrative records of
employees covered by state unemployment insurance tax records. Hours data from establishments provide
consistency with output data in the reporting and coding on industries and thus are well suited for producing
industry-level measures. CES data on employment and average weekly hours paid for production workers in goodsproducing industries and nonsupervisory workers in service-producing industries are the building blocks of labor
input.
69
The hours worked to hours paid ratio is constructed using information from the National Compensation Survey
program; prior to 2000, the annual Hours at Work Survey was used.
70
In August 2004, BLS introduced this new method of constructing estimates of hours for nonproduction and
supervisory workers; see Eldridge, Manser, and Otto. (2004).
71
Employment counts for employees in agricultural services, forestry, and fishing are reported from the BLS’s 202
program, based on administrative records from the unemployment insurance system.
72
Additional information concerning data sources and methods of measuring labor composition can be found at
www.bls.gov/mfp/mprlabor.pdf and in BLS Bulletin 2426 Labor Composition and U.S. Productivity Growth, 194890 (December 1993).
73
The BLS is currently investigating the possibility of constructing labor composition estimates for the
manufacturing sector productivity measures.
145
Capital Inputs
Capital inputs for private business and manufacturing MFP measures are similar. The BLS
capital input measures include assets that are owned and operated by a business within the
sector; rented capital services are included in intermediate inputs. Capital input measures the
services derived from the stock of physical assets and software. The assets included are fixed
business equipment, structures, inventories, and land. Financial assets are excluded from capital input
measures, as are owner-occupied residential structures. The aggregate capital input measures are
obtained by Tornqvist aggregation of the capital stocks for each asset type within each of 60 NAICS
industry groupings using estimated rental prices for each asset type. Rental prices reflect the
nominal rates of return and rates of economic depreciation and revaluation for the specific asset
types. Rental prices are adjusted for the effects of taxes; rental prices of capital are computed for
18 3-digit NAICS industries within manufacturing. Data on investments in physical assets are
obtained from the BEA (see BLS [2009]).
Energy, Materials and Purchased Business Services
In the manufacturing sector, inputs include intermediate inputs, as well as capital and labor
inputs. Intermediate inputs (energy, materials, and purchased business services) are obtained
from BEA’s annual input-output tables. Tornqvist indexes of each of these three input classes are
derived at the 3-digit NAICS level and then aggregated to total manufacturing. For
manufacturing, materials inputs are adjusted to exclude transactions between manufacturing
establishments to maintain consistency with the sectoral output concept. 74
Nominal values of materials, fuels, and electricity and quantities of electricity consumed are
obtained from economic censuses and annual surveys of the Bureau of the Census, U.S.
Department of Commerce. Purchased business services are estimated using benchmark inputoutput tables and other annual industry data from the BEA. Prices of many service inputs are
based on the BLS price programs and are obtained from the national income and product
accounts.
Imported Intermediate Inputs
The BEA produces import matrices as supplementary tables to the annual input-output (I-O)
accounts. For each commodity, the import-matrix table shows the value of imports of that same
commodity used by each industry. Because such information is not available from most
businesses, the estimates must be imputed from data available in the annual I-O accounts. The
imputed-import values are based on the assumption that each industry uses imports of a
commodity in the same proportion as imports-to-domestic supply of the same commodity.
(Domestic supply represents the total amount of a commodity available for consumption within
the United States; it equals domestic output plus imports less exports.) The implication of using
this assumption to calculate the estimates is that all variability of import usage across industries
74
A nonprofit adjustment is made to intermediate inputs, but not to imported intermediates because it is doubtful
that nonprofits are using a significant amount of imported intermediates. By not making a nonprofit adjustment to
imported intermediates, we may overstate the importance of imports slightly.
146
reflects the assumption and is not based on industry-specific information (Strassner, Yuskavage,
and Lee 2009).75
The BEA provided these detailed statistics to the BLS for this research study. These data are not
included in the published tables because their quality is significantly less than that of the higher
level aggregates in which they are included. Compared to these aggregates, the more detailed
statistics are more likely to be either based on judgmental trends, on trends in the higher level
aggregate, or on less reliable source data.76
Using this dataset we can observe trends in the shares of imported intermediate inputs. The share
of intermediate inputs, used by all private industries, that is accounted for by imports grew from
7.6 percent in 1998 to almost 10 percent in 2006. Notice in Figure 1 that there was a decline in
the share of imports used by private industries around the 2001 recession; however, beginning in
2002, they increased steadily. Purchased materials account for the majority of imported
intermediates, and grew steadily, again with a slight dip around the 2001 recession. Imported
material inputs accounted for 15 percent of total materials used by private industries in 1998 and
grew to 21 percent by 2006.77
Although it was once thought that services were not off-shorable, we see evidence that service
inputs are also being imported. Imported service inputs accounted for 1.4 percent of total
intermediates used by private industries in 1998 and 1.7 percent in 2006. However, imported
services inputs account for roughly 3 percent of all service inputs used by private industries, and
this stayed relatively steady from 1998 to 2006. Interestingly, we observe growth in the share of
energy inputs that are imported; 4 percent of all energy inputs used by private industries were
imported in 1998, and we see 12 percent imported by 2006.78 However, imported energy inputs
are less than 0.4 percent of total intermediates used by the private industries.
75
This study uses BEA international transaction account data to assess the import comparability assumption. They
find that real imported materials may be understated in the annual I-O accounts. However, they indicate that the
comparability assumption provides reasonable results at the aggregate level. Feenstra and Jensen(2009) prepare
alternative imported intermediates using an alternative method for allocating imported input across industries and
compare the results with the BEA import matrix that uses the comparability assumption. They find that there are
differences between the two approaches, and identify cells in the I-O table where the differences are greatest.
Unfortunately, data limitations prevent them from resolving the differences
76
Notes about the imported intermediate input data are from BEA documentation that accompanied the data.
77
Imported materials inputs include crude petroleum as a raw material for the refining and coal products industry.
The increase in crude petroleum prices over this time period could be responsible for the increase in imported
materials share of intermediate inputs used by private industries, and more significantly the increase in imported
materials share of intermediate inputs in the manufacturing sector.
78
Crude oil is classified as a nonenergy material input to U.S. refineries, rather than an energy input.
147
Figure 1. Imported Intermediate Inputs Share of Total Intermediates,
by type of input, private industries, 1998-2006
Source: Bureau of Economic Analysis
Looking at the imported intermediate data by industry, we see that the manufacturing sector
consumes over 60 percent of all imported intermediate used by private industries. For the
manufacturing sector, we observe that the share of intermediate inputs that is accounted for by
imports is significantly larger than it is for all private industries and grew at a faster rate. Figure
2 shows imported intermediates share of “sectoral” intermediate inputs (total intermediates less
domestically manufactured inputs), as well as the imports’ share of total intermediates. The
“sectoral” intermediate inputs for the BLS manufacturing sector are less than the total
intermediates in the BEA annual I-O accounts because intermediates that are purchased from
other firms within the U.S. manufacturing sector have been removed. Therefore, the imports’
share of “sectoral” intermediates is greater than the imports’ share of total intermediate inputs.
The “sectoral” intermediate inputs for the manufacturing sector are 57 percent of the BEA total
intermediates.
We observe that 24 percent of “sectoral” intermediates were imported in 1997; this grew to
almost 35 percent in 2006. Notice in Figure 2 that beginning in 2002, there has been a steady
increase in the share of imported intermediates used by U.S. manufacturing firms relative to
“sectoral” and total intermediates.79 As we observed for the private business sector, imported
materials account for the majority of imported intermediate inputs. However, service inputs
were also imported by the manufacturing sector. Imported services’ share of “sectoral”
intermediates in the manufacturing sector grew from 1.3 percent in 1997 to 2.1 percent in 2006,
while imported energy’s share grew from 0.1 percent in 1997 to 0.3 percent in 2006.
79
In 2006, total materials imported by the petroleum industry accounted for 34 percent of material imports by the
manufacturing sector. Over the 1997–2006 period, the price of imported intermediates for the petroleum industry
grew 14 percent as compared to average growth of prices in the manufacturing sector as a whole of 4 percent.
148
Figure 2. Imports Share of Sectoral Intermediate Inputs,
by type of input, U.S. Manufacturing, 1997–2006
(imports’ share of total intermediates also shown)
35%
30%
25%
20%
15%
10%
5%
0%
1997
1998
1999
Imported Energy
2000
2001
Imported Materials
2002
2003
2004
Imported Services
2005
2006
Share of Total
Intermediates
Source: Bureau of Labor Statistics and Bureau of Economic Analysis
BLS MULTIFACTOR PRODUCTIVITY
Solow Model of Productivity
It is generally acknowledged that technical progress can best be captured using a total factor
productivity concept. The most common model of total factor productivity is credited to Solow
(1957). The Solow residual model evaluates technical progress as the difference between the
time derivative of production and the weighted aggregate of the time derivatives for all factors of
production. This measure of disembodied technological change evaluates the ability to expand
the production possibilities frontier without the addition of resources. Given a production
function Y = f ( X ,t ) , the growth rate of total factor productivity, A, can be written as
(1)
•
•
⎛ • ⎞
A Y
Xi
= − ∑ ⎜ ωi ⎟
⎜
A Y
Xi ⎟
i
⎝
⎠
where Δ represents a time derivative , Y denotes real aggregate output, Xi denotes the ith factor
of production, and βi represents the corresponding output elasticity. This productivity growth
model requires well-defined concepts of output and inputs that correspond to a specified
production process. To construct measures of productivity, we must make a discrete
approximation for the time derivatives (Diewert 1976), and we must assume cost-minimizing
behavior in order to measure the βi with cost shares.
149
BLS Multifactor Productivity for the Private Business Sector
The BLS labor productivity measures for the private business sector compare output, measured
as the real gross domestic product of U.S. businesses, to hours worked by all U.S. workers who
contribute to the production of this output. Real GDP is measured by adding all exports and
subtracting all imports from domestic final demand. Thus, imported intermediate inputs are
excluded from the scope of the output measures, and as a result, the contribution of the labor
hours worked overseas that produce the imported intermediate inputs are also absent from the
analysis of U.S. productivity. The output measure used to construct the productivity measure for
the private business sector removes the output of intermediate inputs produced and used within a
sector, as well as all imported intermediate inputs and other domestic intermediate inputs
produced outside the sector. Thus, BLS MFP, ABLS, contains only two factor inputs, labor (L)
and capital services (K), and can be written as
(2)
ΔA
A
BLS
=
ΔY
BLS
Y
BLS
− wL
BLS
ΔL
L
− wK
ΔK
K
Or
(3)
d ln A
BLS
=
d ln Y
BLS
− wL
d ln L − w d ln K
K
where the YBLS is the BLS real private business sector output, dlnABLS denotes the difference in
logarithms of ABLS for successive years (lnA(BLS,t) – lnA(BLS,t-1) ) , and the weights for labor and
N
capital, wi, are the averages of each factor’s nominal cost (Ci) relative to nominal output, Y BLS,
in two successive years:
(4)
wi = L, K
N ⎞
⎛ N
⎜ C i,t C i,t −1 ⎟
= 1/ 2 * ⎜ N + N
⎜ Y BLS,t Y BLS,t −1 ⎟⎟
⎠
⎝
Because of this design it is impossible to observe the impact of offshoring intermediate inputs on
production. To incorporate intermediate inputs into the model, we need to use a sectoral output
concept.
Private Business Sector Multifactor Productivity Adjusted to Include Imports
Sectoral output removes from the value of output only those intermediate inputs that are
produced elsewhere within the sector to eliminate double counting. Intermediate inputs, which
are produced outside of the sector, (i.e., imported intermediates) remain in output (Domar 1961).
To bring imported intermediate inputs inside the major sector model framework, we must not
exclude them as a component of output, and they must be included as a factor input to
production. Denoting the imported intermediate inputs as II, the production function becomes
YS = f ( L, K, II, t). Using this output concept, we can write MFP as:
150
(5)
d ln A
=
S
d ln Y − θ w d ln L − θ w d ln K − ∑ wj d ln IIj
L
S
K
j
⎡
⎢
⎢
⎣
⎤
⎥
⎥
⎦
where the factor weights for imported intermediate inputs of energy (IE), materials (IM), and
services (IS) are defined as
(6)
w
( j = IE ,IM ,IS )
⎛ N
⎜ C j ,t
= 1 / 2* ⎜
+
N
⎜ Y S ,t
⎝
CN
Y
j ,t −1
N
S ,t −1
⎞
⎟
⎟
⎟
⎠
and an output adjustment ratio, θ, used to correct the weights on labor and capital, is written as a
two-period average
(7)
θ
N
⎛ N
⎞
Y
Y
BLS ,t −1 ⎟
BLS ,t
⎜
= 1 / 2*
+
N
N
⎜
Y S ,t −1 ⎟⎠
⎝ Y S ,t
Algebraically working through the model, we derive an adjusted MFP measure that encompasses
imported intermediate inputs in both the output and input indexes. Assuming that growth in
sectoral output is a weighted average of growth in the BLS output measure and intermediate
imports, the resulting MFP growth rate is a scalar of the existing BLS MFP growth:
(8)
d ln A
S
= θ
d ln A
BLS
Table 1. Growth of Components of Private Business Sector Multifactor Productivity,
Alternative Output Concepts, 1997–2006
Original
Output
Sectoral
Output
Labor
Capital
Imported
Intermediates
Imported
Energy
Imported
Materials
Imported
Services
annual growth from previous year
1998
4.9%
5.3%
2.3%
6.3%
10.7%
3.8%
10.9%
10.3%
1999
5.2%
5.4%
2.7%
6.5%
8.5%
9.2%
8.3%
9.3%
2000
3.9%
4.4%
1.0%
6.3%
9.6%
11.2%
9.5%
9.7%
2001
0.5%
0.1%
-1.4%
4.6%
-3.8%
-1.9%
-5.4%
3.8%
2002
1.5%
1.4%
-1.4%
2.9%
-0.1%
-6.5%
-1.3%
5.5%
2003
3.1%
3.1%
-0.3%
2.3%
3.1%
3.4%
4.4%
-2.4%
2004
4.3%
4.9%
1.5%
2.3%
11.8%
27.3%
10.3%
16.4%
2005
3.7%
3.9%
1.8%
2.5%
5.7%
13.9%
5.6%
4.7%
2006
3.2%
3.4%
2.6%
2.7%
4.9%
2.8%
4.7%
6.8%
3.5%
1.0%
4.0%
5.5%
6.6%
5.1%
7.0%
average annual growth
19972006
3.4%
151
Table 1 presents growth rates for the components of the MFP model for the private business
sector.80 Notice that the imported intermediates grow faster than labor and capital in most years,
except around the 2001 recession. The growth of imported intermediate inputs has an impact on
the growth of sectoral output trends as well, which grew somewhat faster than the published
value-added output measure for all years except 2001 and 2002. The year-to-year growth rates
of the imported intermediates fluctuate quite a bit. Over the 1997–2006 period, energy and
service imports grew faster than imported materials. However due to their small size, imported
materials growth is driving the growth in total imported intermediate inputs.
Using BEA estimates of imported intermediate inputs, we derived the adjustment scalar for the
private business sector MFP measures. Table 2 shows the results of adjusting the published BLS
MFP data. Notice that by incorporating the imported intermediate inputs into the MFP
framework, the annual growth in private business sector MFP is reduced by 0.1–0.2 percentage
points.
Table 2. Multifactor Productivity Growth for the Private Business Sector,
by alternative treatment of imports, 1997–2006
Excluding Imported
Intermediate Inputs:
Including Imported
Intermediate Inputs
Difference
1.30%
1.29%
1.28%
0.11%
1.65%
2.63%
2.49%
1.63%
1.20%
1.19%
1.18%
0.10%
1.53%
2.43%
2.28%
1.48%
-0.10%
-0.10%
-0.10%
-0.01%
-0.13%
-0.20%
-0.20%
-0.15%
0.54%
0.49%
-0.05%
1.43%
1.32%
-0.12%
BLS published data
annual growth from previous year
1998
1999
2000
2001
2002
2003
2004
2005
2006
annual average growth
1997-2006
Substitution of Imported Intermediates for U.S. Labor in the Private Business Sector
Using the Solow MFP equation, we estimated the effects of substitution between imported
intermediate inputs and U.S. hours worked on labor productivity. The growth in imported
intermediate inputs, combined with growth in capital inputs and technical change, directly
80
The time series does not cover the business cycles sufficiently to divide that data into subperiods that would allow
a meaningful analysis of the data. We constructed subperiods of 1997–2000 and 2001–2006, as well as 1997–2002
and 2003–2006. The comparison of results between period 1 and period 2 was very sensitive to the year that the data
was divided. Therefore, we will not present subperiod analysis in this paper.
152
influence labor productivity. Thus, labor productivity can be written as the sum the intensity of
each of the other input factors (increases in the factor’s quantities relative to domestically
employed labor):
(9)
d lnY − d ln L = d ln A +θ w d lnK d ln L + ∑ wj ⎛⎜⎝ d ln IIj − d ln L ⎞⎟⎠
S
S
⎛
⎜
K ⎜
⎝
−
⎞
⎟
⎟
⎠
j
Figure 3 shows the contributions to private business sector labor productivity of the remaining
nonlabor factor inputs. From 1997 through 2002, growth in capital services contributed to the
majority of labor productivity growth. Beginning in 2003, capital’s contribution to labor
productivity declined and was outpaced by MFP growth. Also, beginning in 2004 the
contribution of imported intermediate inputs contributed more to labor productivity growth than
capital growth. Again, we note that the influence of imported material inputs dominated the
contribution of all imported intermediate inputs.
Figure 3. Labor Productivity Growth by Contributing Input Factors,
Private Business Sector, 1997–2006
(annual growth rates from the previous year)
Sources: Bureau of Labor Statistics (using BEA unpublished import data)
Using the sectoral output approach, we observe (see Table 3) that for the 1997–2006 period,
approximately 14 percent of labor productivity growth was attributed to growth in imported
intermediate inputs (11 percent to materials, 3 percent to services, and less than 0.5 percent to
energy).81
81
Note that because output has been expanded to include imports, the labor productivity growth is 2.6 percent per
year, rather than 2.4 percent per year.
153
Table 3. Labor Productivity Growth and the Contribution of Non-labor Inputs and
Multifactor Productivity, U.S. Private Business Sector 1997–2006
(average annual growth rates)
Output per unit of labor (includes imports)
2.56%
Multifactor Productivity (includes imports)
1.31%
Contribution of capital intensity
0.88%
Contribution of imported intermediates
0.37%
Contribution of imported materials
0.27%
Contribution of imported services
0.08%
Contribution of imported energy
0.01%
BLS MULTIFACTOR PRODUCTIVITY FOR THE U.S. MANUFACTURING SECTOR
As mentioned earlier, BLS productivity measures for the manufacturing sector are constructed
using a sectoral output concept. Therefore, imported intermediates are within the productivity
model framework. For the MFP measures, imported intermediate inputs are a component of
measured output and intermediate inputs. To identify the impact of imported intermediates on
manufacturing productivity, we do not need to adjust the measures to include imports, but rather
separate the intermediates into domestic and imported components. This demarcation is achieved
using the BEA estimates of imported intermediates, which were provided to the BLS at the
industry level of detail.
Table 4 presents the year-to-year growth rates and the average annual growth for the components
of the manufacturing MFP model over the 1997–2006 period. Notice that in most years, labor
inputs declined and imported intermediates grew faster than capital and domestic
nonmanufactured intermediate inputs. Prior to the 2001 recession, there was strong growth in
capital services, imported intermediates, and domestic nonmanufactured intermediates. However,
note that domestic nonmanufactured intermediates were impacted by the recession sooner than
the imported intermediates. Also notice that the imported intermediates were able to rebound
after the recession, while domestic nonmanufactured inputs showed negative growth through
2004. Over the entire 1997–2006 period, labor and domestic nonmanufactured intermediates
inputs declined, while capital services and imported intermediates grew.82
82
Kurz and Lengermann (2008) construct a gross output productivity measure in order to keep U.S. manufactured
intermediates in the model. This model allows an analysis of the shift from domestic to imported intermediate
inputs.
154
Table 4. U.S. Manufacturing Sector Multifactor Productivity and Components, 1997–2006
Sectoral
Output
Labor
Capital
Domestic
Intermediates
Imported
Intermediates
MFP
5.0%
4.1%
3.1%
1.5%
0.6%
0.0%
-0.6%
0.0%
0.5%
2.3%
4.2%
-4.1%
-3.0%
-4.4%
-1.3%
-5.2%
7.7%
-2.0%
9.6%
7.1%
5.5%
-4.9%
-2.1%
2.6%
8.7%
4.9%
4.3%
2.30%
0.80%
3.50%
-1.30%
3.70%
2.80%
2.60%
0.40%
1.60%
1.57%
-0.74%
3.88%
1.79%
Annual growth from previous year
1998
1999
2000
2001
2002
2003
2004
2005
2006
5.2%
3.8%
2.7%
-5.1%
-0.7%
1.0%
1.7%
3.7%
1.8%
-0.2%
-0.7%
-1.3%
-6.5%
-7.1%
-4.9%
-0.5%
-1.1%
0.6%
Annual average growth
19972006
1.53%
-2.44%
* Combined intermediates constructed as a weighted aggregate of energy, materials, and purchased services.
Table 5 compares the growth of domestic nonmanufactured intermediate inputs and imported
intermediates by type of input. In general, we note that imported intermediates showed stronger
growth than domestically nonmanufactured inputs. It is interesting to note that domestic material
inputs (excluding materials purchased from other manufacturing industries) were declining in
most years, while imported materials grew.
155
Table 5. Comparison of Imported and Domestic Intermediate Inputs by Type of Input, U.S.
Manufacturing Sector, 1997–2006
Total Intermediates
ENERGY
Domestic Imported Domestic
Annual growth from previous year
MATERIALS
Imported
Domestic
Imported
SERVICES
Domestic
Imported
1998
2.25%
9.59%
-2.49%
-7.80%
1.94%
9.73%
3.02%
8.48%
1999
4.21%
7.12%
0.09%
0.39%
3.79%
6.57%
4.93%
15.76%
2000
-4.10%
5.52%
-5.04%
-11.12%
-10.12%
5.85%
-0.06%
1.54%
2001
-3.02%
-4.86%
-9.47%
-6.99%
-6.13%
-7.29%
-0.48%
28.48%
2002
-4.44%
-2.11%
-1.51%
-1.17%
-8.39%
-2.14%
-2.53%
-1.82%
2003
-1.25%
2.64%
-6.08%
12.96%
-4.87%
3.17%
1.14%
-4.19%
2004
-5.23%
8.71%
-2.15%
35.05%
-9.97%
8.12%
-2.89%
13.88%
2005
7.74%
4.93%
8.05%
25.06%
7.44%
4.63%
7.87%
6.38%
-2.02%
4.25%
Average annual growth
-6.81%
10.69%
-7.40%
3.91%
1.67%
8.20%
-2.94%
5.34%
-3.93%
3.49%
1.36%
8.13%
2006
19972006
-0.74%
3.88%
*Combined intermediates constructed as a weighted aggregate of energy, materials, and purchased services
Figure 4 presents the trends in constant-dollar factor input costs for the U.S. manufacturing
sector. Notice that labor represents the highest cost and was constant prior to the 2001 recession,
when it declined with falling employment in manufacturing. Energy and imported services
represent a very small portion of the overall factor costs in manufacturing and were relatively
constant over the 1997–2006 period. Interestingly, the cost of imported materials increased over
the period, while the cost of domestic nonmanufactured materials declined. The factor costs of
capital services and purchased domestic services increased somewhat. We next estimate the
effects of imported intermediate inputs on labor productivity by using the Solow MFP model.
156
Figure 4: Input Costs for the Manufacturing Sector, by type 1998–2006
Constant dollar, billions
Substitution of Imported Intermediates for U.S. Labor in the Manufacturing Sector
The model used by BLS to measure MFP for the U.S. manufacturing sector can be written as
(10)
d ln A
G
− wE
=
d ln Y
G
− wL
d ln L − w d ln K
K
d ln E − w d ln M − w d ln S
M
S
where YG is real sectoral output for the manufacturing sector, dlnAG denotes the difference in
logarithms of AG for successive years (lnA(G,t) – lnA(G,t-1) ), and the weights for labor, capital,
energy, materials, and purchased business services, wi, are the averages of each factor’s nominal
N
cost relative to nominal output, Y G in two successive years:
(11)
wi = L, K , E , M , S
⎛C
C i ,t −1 ⎞
i ,t
= 1/ 2 * ⎜ N + N ⎟
⎜
⎟
⎝ Y G ,t Y G ,t −1 ⎠
157
The growth in imported intermediate inputs, combined with growth in capital inputs, domestic
intermediate inputs, and technical change, directly influence labor productivity. Thus, labor
productivity can be written as the sum the intensity of each of the other input factors (increases
in the factor’s quantities relative to domestically employed labor):
(12)
d lnY
−
G
∑ wDI
j
j
⎛⎜
⎝
⎛
⎜
K ⎜⎜
⎝
d lnK d ln L
+
∑ wII
d ln L = d ln A + w
G
d lnDIj − d ln L ⎞⎟⎠
−
j
j
⎛⎜
⎝
⎞
⎟
⎟⎟
⎠
+
d lnIIj − d ln L ⎞⎟⎠
where wDIj denotes the weights on domestic intermediates j = E,M,S and wIIj denotes the
weights on imported intermediates j = E,M,S.
Figure 5. Labor Productivity Growth by Contributing Input Factors,
Manufacturing Sector, 1998–2006
(annual growth rates from previous year)
Figure 5 shows the contributions of nonlabor factor inputs to year-to-year growth of
manufacturing sector labor productivity, and Table 6 presents the contributions of nonlabor
factor inputs on the average annual growth over the entire period from 1997 to 2006. From
Figure 5, notice that in most years, MFP contributed the most to labor productivity growth. Also
notice that growth in capital services contributed to labor productivity growth prior to 2004, but
very little thereafter. Imported intermediate inputs made a relatively constant contribution to
labor productivity growth in all years, with the exception of 2001. Over the period 1997–2006,
MFP accounted for 45 percent of productivity growth and imported intermediate inputs
accounted for 23 percent.
158
Table 6. Contributions to Labor Productivity in the U.S. Manufacturing Sector 1997–2006
(average annual growth)
Output per unit of labor
3.96%
Multifactor Productivity
1.79%
Contribution of capital intensity
0.64%
Contribution of domestic intermediates
0.65%
Contribution of imported intermediates
0.92%
Contribution of imported materials
Contribution of imported services
Contribution of imported energy
0.80%
0.10%
0.01%
INFLUENCE OF IMPORT PRICES
To assess the impact of possible bias in the price change of imports on productivity, we consider
the difference between the growth of the BLS productivity measure, dln ABLS, and the growth of
a productivity measure that is constructed with more precise price indexes for imports, dln
Aprice*. Prices of imports enter the BLS private business sector productivity model when imports
are removed from final demand in the construction of real GDP (which is further reduced to
arrive at private business sector output, YBLS). To assess the impact of possible import price bias,
we assume that domestic inputs and all other components of output are measured precisely. 83
Therefore, the possible bias in productivity growth equates to a difference in the growth of
alternative output measures:
(13)
d ln A
BLS
−d
ln A
Pr ice*
=
d ln Y
BLS
−d
ln Y
Pr ice*
By assuming that all domestic components of output are measured precisely, the difference in the
growth of measured output and an output measure that is constructed using alternate import
prices becomes the difference in the growth of measured imports, IBEA, and the alternate, IPrice*,
that is measured with alternative import prices. The growth of the differences in import measures
must be weighted by imports’ share, sI, of output. Because the shares are calculated using
nominal data, there is no difference in the weights. The difference in the growth of measured
productivity relative to a productivity measures constructed with alternative prices of imports
becomes
83
Diewert and Nakamura (2009) present a new measure for the bias in an import price index due to outsourcing and
show how the price index bias problems attributable to input source substitution can be represented theoretically.
The authors suggest that outsourcing and the inability of price indexes to capture the effect of new firm entry and
input substitution, have led to an upward bias in intermediate input and import price indexes.
159
(14)
d ln A
BLS
−d
ln A
Pr ice*
= − s I ⎛⎜
⎝
d ln I
BEA
− d ln
I
Pr ice*
⎞⎟
⎠
Where
N
⎛ N
⎞
⎜ I t
I
t −1 ⎟
= 1 / 2* ⎜ N + N
⎟
⎜ Y BLS,t Y BLS ,t −1 ⎟
⎝
⎠
sI
(15)
Real growth in imports can be calculated as the difference between nominal growth and price
growth. As there is no difference in nominal growth between the two concepts, the difference
between the growth of measured productivity and a productivity measure constructed with
alternative import prices becomes the weighted difference between the measured price growth of
I
I
imports, P BEA, and an alternative measure of price growth, P Price*
(16)
d ln A
BLS
− d ln
A
Pr ice*
=
s
I
⎛⎜
⎝
d ln P I
BEA
− d ln
P
I
Pr ice*
⎞⎟
⎠
Note that the value of aggregate imports is based upon many individual commodities that may or
may not suffer from biased import prices. An individual commodity’s impact on aggregate
I
productivity growth will be determined by the bias in that commodity’s price, P i, growth,
weighted by the imported commodity’s share of output , ci:
(17)
N
⎛ N
⎞
I
I
i ,t −1 ⎟
i ,t
⎜
= 1 / 2*
+
N
⎜ N
⎟
⎝ Y BLS ,t Y BLS ,t −1 ⎠
I
ci
An individual commodity’s impact on productivity growth can be estimated as
(18)
c
I
i
⎛⎜
⎝
d ln P I
BEA ,i
− d ln
PI
Pr ice*,i
⎞⎟
⎠
The size of the possible bias in aggregate productivity growth is
(19)
∑c
i
I
i
⎛⎜
⎝
d ln P I
BEA ,i
−d
ln P I
Pr ice*,i
⎞⎟
⎠
When we modified the BLS private sector MFP model to include intermediate inputs in the
model, we reduced the influence of possible price bias on the output component; however we
introduced the possible price bias on the input side of the model. Again we consider the impact
of import prices on productivity as the difference between the growth of the modified
productivity measure, dln AS, and the growth of a productivity measure that is constructed with
more precise price indexes for imports, dln Aprice*. Recalling the modified MFP Eq. (5), the
160
difference between the growth of modified productivity and a productivity measure constructed
with alternative import prices is the difference in output growth and the weighted difference the
growth of imported intermediate inputs with existing import price indexes, II BEA, and an
alternative measure of price growth, IIPrice*
(20)
d ln A
− d ln
S
A
Pr ice*
d ln Y
=
S
− d ln Y Pr ice* − w II ⎛⎜
⎝
d ln II
BEA
− d ln
II
Pr ice*
⎞⎟
⎠
In the modified MFP model, only imports used as intermediate inputs in production are added
back into the model. Assuming they can be added back in the same manner as they were
originally removed, the growth of real output can only be biased to the extent that price measures
for imports that are destined for final demand are biased. The possible impact on productivity
growth is estimated as the weighted difference between the measured price growth of imports,
PIBEA, and an alternative measure of price growth, PIPrice* over final demand products and
intermediate inputs
final _ demand
∑ w ⎛⎜⎝ d ln P I
i
(21)
− d ln
P
I
BEA,i
− d ln
P
I
BEA,i
I
i
int ermediates
∑ w I ⎛⎜⎝ d ln P I
i
i
Pr ice*,i
⎞⎟ +
⎠
Pr ice*,i .
⎞⎟
⎠
Because the weights on the final demand components and the intermediate inputs are both that
commodity’s share of nominal output, the influence of mismeasured import prices on aggregate
productivity is
I
w
(22)
⎛⎜
⎝
d ln P I
BEA
− d ln
PI
Pr ice*
⎞⎟
⎠
where
(23)
w
I
=
N
⎛ N
⎞
I
I
⎜
t −1 ⎟
t
+
1/ 2 *
N
⎜ N
⎟
⎝ Y S ,t Y S ,t −1 ⎠
By construction, wj is less than sj for all commodities; recall that YS > YBLS . Therefore, the
impact of import prices on MFP is smaller under the modified MFP framework than in the BLS
published MFP model.
161
Table 7: Imported Intermediate Inputs Share of Private Business Sector Output,
1998–2006
BLS Output Share, sI
Sectoral Output Share,
wI
1998
1999
2000
2001
2002
2003
2004
2005
2006
8.05%
8.07%
8.76%
8.84%
8.25%
8.25%
8.98%
10.03%
10.77%
7.45%
7.47%
8.05%
8.12%
7.62%
7.62%
8.23%
9.11%
9.72%
Because import prices are not used to construct real output measures for the BLS manufacturing
productivity statistics, any possible price mismeasurement of imports does not affect labor
productivity statistics for the manufacturing sector. However, prices of imports enter the BLS
manufacturing sector MFP model when imports are included in the construction of purchased
intermediate inputs. To assess the impact of possible import price bias, we assume that output
and all domestic inputs are measured precisely. Therefore, the possible bias in productivity
growth equates to a difference in the weighted growth of imported intermediate inputs
d ln A
− d ln
S
(24)
(25)
A
Pr ice*
= − wi ⎛⎜
⎝
d ln II
BEA
− d ln
II
Pr ice*
⎞⎟
⎠
⎛
I
I ⎞
= 1 / 2 * ⎜ iN,t + iN,t −1 ⎟
⎜
⎟
⎝ Y G ,t Y G ,t −1 ⎠
wi = IE ,IM ,IS
Real growth in imports is calculated as the difference between nominal growth and price growth.
As there is no difference in nominal growth between the two concepts, the difference between
the growth of measured productivity and a productivity measure constructed with alternative
import prices becomes the weighted difference between the measured price growth of imports,
PIBEA, and an alternative measure of price growth, PIPrice*
I
(26)
∑w
i
I
j
⎛
⎜
⎝
d ln P I
BEA, j
−d
ln P I
162
Pr ice*, j
⎞
⎟
⎠
Table 8: Imported Intermediate Factor Cost Shares, Manufacturing Sector, 1998-2006
BLS Output Share,
wI
1998
1999
2000
2001
2002
2003
2004
2005
2006
12.24%
12.39%
13.53%
13.97%
13.57%
13.86%
15.24%
16.94%
18.33%
CONCLUSIONS
In this paper we develop a framework for estimating the effects of imported intermediate inputs
on U.S. major sector labor productivity. The production model used to calculate the BLS private
business sector MFP measures is expanded to treat imported intermediate inputs as an input
rather than as a subtraction from output. Once the imported intermediate inputs are inside the
framework, we use the Solow MFP equation to estimate the effects on labor productivity of
substitution between imported intermediate inputs and U.S. hours worked. Separate effects are
estimated for imported energy, materials, and services. The data show that imports increased as
a share of total intermediates used by private industries from 8 percent in 1997 to 10 percent in
2006. By including imported intermediates in the MFP model, we find that private business
sector MFP grew 0.1–0.2 percent per year slower than the BLS published series. Also, we
estimated that the growth in imported intermediate inputs contributed 14 percent to the average
annual growth of labor productivity for the private business sector from 1997 to 2006.
We do not believe that it would be a good idea to alter the labor productivity model to
incorporate imported intermediates, as then the trend could be considered “biased” to the extent
that output would reflect the growth in imported intermediates, while the labor input would not
include the corresponding hours worked overseas. However, the role of imported intermediates
can be meaningfully assessed in the MFP model. From the exercise above (see Table 2), we find
that including imported intermediates in a sector output concept and as a factor input in
production, MFP grew 0.1–0.2 percent per year slower than the BLS published series.
Because over 60 percent of imported intermediate inputs purchased by private industries are used
by the manufacturing sector, we also evaluate the role of imported intermediates in the U.S.
manufacturing sector. The BLS methods for constructing manufacturing MFP include
intermediates in the model framework. Therefore, we isolate the imported components to assess
their impact on labor productivity. The data reveal that over the 1997–2006 period, imported
intermediate inputs grew as a share of total intermediate inputs. We find that labor inputs and
domestic nonmanufactured inputs declined over the entire period, while capital services and
imported intermediates grew. In addition, we estimate that growth in imported intermediate
inputs contributed 23 percent to the average annual growth in labor productivity in the
manufacturing sector.
Finally, we show that any mismeasurement of import prices impacts BLS productivity measures.
However, the impact will be weighted by the share of imports relative to aggregate output, which
ranged from 8–12 percent for the private business sector and 12–18 percent for the
manufacturing sector.
163
References
Bureau of Labor Statistics (BLS). 2009. “Multifactor Productivity Trends, 2007.” News release.
Washington. DC: U.S. Department of Labor, #09-0302, March 25.
Diewert, W. Erwin. 1976. “Exact and Superlative Index Numbers.” Journal of Econometrics 4:
15–145.
Diewert. W. Erwin, and Alice O. and Nakamura. 2009. “Bias in the Import Price Index Due to
Outsourcing: Can It Be Measured?” Paper prepared for the conference “Measurement
Issues Arising from the Growth of Globalization,” held in Washington, DC, November
6–7.
Domar. Esvey. 1961. “On the Measurement of Technological Change.” Economic Journal 71:
709–729.
Eldridge, Lucy P., Marilyn E. Manser, and Phyllis F. Otto. 2004. “Alternative Measures of
Supervisory Employee Hours and Productivity Growth.” Monthly Labor Review 127(4):
9–28.
Feenstra, Robert C., and J. Bradford Jensen. 2009. “Evaluating Estimates of Materials
Offshoring from U.S. Manufacturing” Prepared for the conference “Measurement Issues
Arising from the Growth of Globalization,” held in Washington, DC, November 6–7.
Kurz, Christopher, and Paul Lengermann. 2008. “Outsourcing and U.S. Economic Growth: The
Role of Imported Intermediates.” Paper presented at the 2008 World Congress on
National Accounts, May 12–17, 2008, Washington, DC.
Solow, Robert M. 1957. “Technical Change and the Aggregate Production Function.” Review of
Economics and Statistics 39(August): 312–320.
Strassner, Erich H., Robert E. Yuskavage, and Jennifer Lee. 2009. “Imported Inputs and Industry
Contributions to Economic Growth: An Assessment of Alternative Approaches.”
Prepared for the conference “Measurement Issues Arising from the Growth of
Globalization,” held in Washington, DC, November 6–7.
Young, Allan H., and Helen Stone Tice. 1985. “An Introduction to National Economic
Accounting.” Survey of Current Business 65(3): 59–76.
165
SESSION 4: OFFSHORING AND PRICE
MEASUREMENT: INDUSTRY CASE STUDIES
CHAIR: Jon Steinsson (Columbia University)
Offshoring and Price Measurement in the Semiconductor Industry,
David Byrne (Federal Reserve Board), Brian Kovak (University of Michigan), and
Ryan Michaels (University of Michigan)
Measuring IT Software and Services: Implications for Real Trade Flows, Catherine Mann
(Brandeis University and Peterson Institute for International Economics)
Imports of Intermediate Parts in the Auto Industry—A Case Study, Thomas H. Klier (Federal
Reserve Bank of Chicago) and James M. Rubenstein (University of Miami, Ohio)
DISCUSSANT: Kimberly Zieschang (IMF)
167
Offshoring and Price Measurement in the
Semiconductor Industry*
†
David Byrne
Federal Reserve Board
‡
Brian K. Kovak
University of Michigan
§
Ryan Michaels
University of Michigan
Initial Draft: October 30, 2009
This Draft: April 1, 2010
Preliminary and incomplete
Please do not cite without permission.
Abstract
The recent growth in offshore outsourcing of intermediate input production makes it especially critical
that statistical agencies are able to accurately measure quality-adjusted trade flows. This paper
examines the implications of global production sharing for measuring the price of semiconductors, a
critical input to high-end domestic manufacturing and U.S. productivity growth. We analyze new
transaction-level data on semiconductor wafer fabrication around the world, including prices and
detailed information on key physical attributes of semiconductor wafers. Semiconductor wafers are a
high-value intermediate good in the production of final products in the semiconductor industry. Using
this detailed information on wafer pricing and quantities by country, we study the price measurement
implications of shifts in the pattern of international production sharing for this important stage in the
value chain. We estimate that, after adjusting for changes in product characteristics, the average
annual price decline in processed wafers was roughly 12.5 percent during the last five years. Our
analysis also finds that shifts in the location of production to lower-cost countries can contribute an
additional price decline of up to 0.8 percent per year.
* Paper prepared for the conference “Measurement Issues Arising from the Growth of Globalization,” sponsored by
the W.E. Upjohn Institute for Employment Research and the National Academy of Public Administration. We
would like to thank Susan Houseman, Marshall Reinsdorf, Jodi Shelton, Sonya Wahi-Miller, and Kim Zieschung for
helpful discussions and Chelsea Boone at GSA for all of her help with data and background information on the
industry. Remaining errors are our own. This paper reflects the views of the authors and should not be attributed to
the Board of Governors of the Federal Reserve System or other members of its staff.
†
David Byrne, Federal Reserve Board, Washington, DC
‡
Corresponding author: Brian K. Kovak, PhD Candidate, Department of Economics, University of Michigan. Email:
[email protected]
§
Ryan Michaels, PhD Candidate, Department of Economics, University of Michigan
169
The recent growth in offshore outsourcing of intermediate input production has generated
concern that standard government data collection methods are ill-suited to an increasingly
international productive structure (Houseman 2007). This paper focuses on the semiconductor
industry to estimate the effects of offshore outsourcing on input price measurement. We find
that offshoring in this industry necessitates the collection of very detailed product data to
adequately adjust prices for input quality, and that shifting sourcing patterns may cause standard
price measures to understate price declines for processed semiconductor wafer inputs by as much
as 0.8 percent per year.84
We choose to examine wafer fabrication, an intermediate stage in semiconductor production, for
a number of reasons. First, semiconductor wafer production has moved offshore to a dramatic
degree in the last forty years, with continual shifts in the geographic distribution of
semiconductor manufacturing capacity. Second, China’s entrance in the semiconductor
manufacturing market in 2001 was much heralded in the media, and provides an interesting case
study on the effects of growing Chinese economic strength on an important industry. Third, the
discrete nature of technological progress in semiconductor wafer fabrication techniques makes
careful quality adjustment feasible, as we describe in detail below. Finally, we have obtained a
new dataset of semiconductor input prices with information on country of origin, making
possible an empirical investigation of the effects of shifts in sourcing on input price
measurement.
Offshoring poses a number of challenges for price measurement in the semiconductor
manufacturing sector in particular. First, suppose a U.S.-based manufacturer contracts out all
production to a firm overseas and that, prior to its decision to offshore, it had purchased final
goods from an independent supplier here in the U.S. or had made the good itself. The one-time
decline in the price level associated with the decision to offshore is not captured by current datacollection procedures. The Producer Price Index’s universe does not include imports, so it does
not reflect the price reduction. The Bureau of Labor Statistics (BLS) International Price Program
(IPP) measures price changes beginning in the second month in which the imported good is
observed, as it is not designed to measure the initial price decline that occurs when a domestic
producer first off-shores a segment of production. A similar problem can arise if the firm has
already contracted out production overseas but now sources from a low-cost supplier in China
rather than from a producer in Taiwan.85
The problem posed by shifting sourcing arrangements is essentially equivalent to the problem of
outlet substitution bias in the CPI, described in detail by the Boskin Commission Report (Boskin
et al. 1996) and Diewert (1998). While those studies were concerned with consumers shifting
their consumption toward low-cost retail outlets, this paper confronts the problem of
semiconductor producers shifting their intermediate input sourcing toward low-cost suppliers
located abroad. The bias is most acute whenever the inputs, as in our case, are approximately
84
Semiconductor wafers are described in detail in the next section.
In principle, the IPP would measure this change if the manufacturer imported the good itself or if it continued to
work through the same intermediary that is surveyed by IPP. If, on the other hand, the manufacturer contracts with a
different intermediary in order to access a new market overseas, the IPP will miss the price decline since it surveys
the importer, which in this case was the original intermediary. Unfortunately, to the best of our knowledge, there is
little information on the relative importance of intermediaries in the IPP.
85
171
identical, which implies that the unmeasured price change when production is shifted to a new
location does in fact represent a genuine price decline for the same good.
The final significant challenge is quality adjustment. As a greater share of production is shifted
abroad, the composition of imports becomes increasingly sophisticated. This is particularly true
within the semiconductor industry, which imports many complex intermediate inputs at various
stages in the production process. This process places much greater demands on quality
adjustment procedures for import prices, as semiconductor technology changes so quickly. The
challenge of quality adjustment in the semiconductor industry is well known and has been
demonstrated in many previous studies.86
We address these concerns using new transaction-level data on semiconductor wafer purchases,
collected by the Global Semiconductor Alliance (GSA). These data contain fine detail on
product characteristics, allowing us to generate constant-quality price indexes. They also report
the source country for each transaction, making it possible to examine the effects of shifting
geographic production on price measurement. Our results demonstrate the importance of having
such detailed data when constructing price indexes in industries with large amounts of
offshoring. This need is likely to increase as more countries move up the technical ladder and
begin exporting ever more complex products.
The paper proceeds as follows. The next section describes aspects of the semiconductor
manufacturing process that are relevant to price measurement. We then describe the data we
utilize to build input price measures, followed by a section that presents our price index
calculations. We begin with a standard matched model index as a baseline and then follow
Reinsdorf (1993) to bound the potential effect of outlet substitution bias due to shifting input
sourcing across countries. This section concludes with comparisons to a hedonic index and a
publicly available official semiconductor price index. The last section concludes.
SEMICONDUCTOR PRODUCTION
This section describes the semiconductor manufacturing process and recent changes in the
business models employed by semiconductor firms, highlighting characteristics of the industry
that are important for price measurement. Semiconductor production technology progresses in
distinct measurable steps, allowing us to account for technological improvement when
constructing price indexes in spite of rapid changes over time. The continuing movement to
outsource semiconductor production to offshore firms raises the possibility of outlet-substitution
bias in standard price indexes and motivates our choice to focus on foundry wafer fabrication.
Semiconductor Production Technology
Semiconductor fabrication involves creating networks of transistors on the surface of a thin piece
of semiconducting material.87 The process begins with the design and layout of a new chip.
Semiconductor designers use suites of complex software to specify the functionality of the chip,
86
87
See, among others, Flamm (1993), Grimm(1998), and Aizcorbe, Corrado, and Doms (2003).
Turley (2003) provides an accessible overview of semiconductor technology, manufacturing, and business.
172
convert that logic into the corresponding network of transistors, determine the physical layout of
those transistors, and simulate the behavior of the proposed design for debugging purposes.
Semiconductors are manufactured in a facility called a fab. Transistors are created on the surface
of the wafer through a photolithography process, in which successive layers of conducting and
insulating materials are deposited on the surface of the wafer and chemically etched away in the
appropriate places to form the desired pattern of transistors and necessary interconnections. The
design layout software determines the etching pattern for each layer, which is projected onto the
wafer through a mask containing the negative of the desired pattern, in a process similar to
developing a photograph by projecting light through a negative. Each step of the etching process
is repeated multiple times across the wafer, resulting in a grid pattern of many copies of the chip.
Once all transistors and connection layers are complete, the chips are tested in a process called
“wafer probe,” and any faulty chips are marked to be discarded. The wafer is then cut up,
leaving individual chips, called die. The die are then placed inside protective packages and
connected to metal leads that allow the chip to be connected to other components.
Semiconductor fabrication technology has advanced over time in discrete steps, defined by wafer
size and line width (also called feature size). Increases in wafer size allow larger numbers of
chips to be produced on a wafer. Most fabs currently produce 150mm (roughly 6 inches),
200mm (8 inches), or 300mm (12 inches) diameter wafers. Although larger wafers cost more to
produce, the move to a larger wafer has generally reduced the cost per die by approximately 30
percent per die (Kumar 2007).
Line width is the size of the smallest feature that can be reliably created on the wafer. Decreased
line width means that individual transistors are smaller, and more functionality can be integrated
into a given area of silicon. This makes chips of a given functionality smaller, lighter, and faster,
and also makes it feasible to include more functions on a single chip. The number of transistors
that can be produced on a chip has grown exponentially over time, following Moore’s Law, the
Intel cofounder’s famous observation that the number of transistors on a chip doubled every
eighteen months (Moore 1965).88 Figure 1 shows the maximum number of transistors per chip
and the minimum line width used to produce Intel processors over the last 40 years (both plotted
on logarithmic scales).
Current line widths are measured in microns (μm) or nanometers (nm). The smallest line width
currently being produced in volume is 25nm. As a rule of thumb, Kumar (2007) estimates that
moving a given chip design to a 30 percent smaller line width will result in cost savings of
approximately 40 percent, assuming the same number of defects in both processes. The primary
drawback of smaller line widths is increased cost per wafer, particularly early in the technology’s
life span. Masks are much harder to produce when creating smaller features, and new process
technologies often result in higher defect rates and lower yields, the fraction of chips on a wafer
that function correctly. In spite of these challenges, the benefits of increased die per wafer and
better performance outweigh the problems of decreased yields, particularly as the fabrication
technology matures and yields increase. Given the benefits of smaller line widths,
semiconductor manufacturers have steadily moved toward newer technology. This is apparent in
Figure 1 for Intel processors and can be seen even more clearly in Figure 2, which plots the
88
This regularity later slowed to doubling every two years.
173
technology composition of sales at Taiwan Semiconductor Manufacturing Company (TSMC),
the largest semiconductor foundry.
There are a number of options regarding the chemicals used to create the transistors themselves
and how the transistors are arranged to implement logical functions. The most common
technology, called complementary metal-oxide semiconductor (CMOS), accounted for 97
percent of worldwide semiconductor production in 2008.89 Other transistor arrangements, such as
bipolar logic, and other chemical processes, such as Gallium Arsenide (GaAs) or Silicon
Germanium (SiGe), generally focus on niche markets for high-frequency, high power, or
aerospace devices, rather than the storage and computational logic products comprising the
majority of the CMOS market. In the following analysis, we will refer to each combination of
wafer size, line width, and logic family as a “process technology” (e.g., 200mm, 180nm, CMOS
constitutes one process technology).
The price index calculations below require us to define the set of product characteristics that
determine the performance, and hence the price, of a given wafer. To guide this choice, we have
consulted pricing models used by engineers at fabless firms to estimate production costs when
developing business plans. Kumar (2008) presents a wafer cost model based on wafer size, line
width, and logic family. A commercial cost estimation firm, IC Knowledge, distinguishes wafer
cost estimates by wafer size, line width, logic family, number of polysilicon layers, and number
of metal layers. Given this potential importance of the number of layers in a given design,
indicating the design’s complexity, we calculate price per layer rather than price per wafer.
These pricing models support the use of process technology (wafer size, line width, and logic
family) to distinguish between goods in our price indexes, calculated in the section “Price Index
Results.”
Changing Semiconductor Business Models
In the early 1970s nearly all semiconductor producers were vertically integrated, with design,
wafer fabrication, packaging, testing, and marketing performed within one company. By the
mid-1970s, firms began moving packaging and test operations to East Asia to take advantage of
lower input costs (Scott and Angel 1988, Brown and Linden 2005). In spite of outsourcing these
relatively simple steps in the production process, firms maintained their complex wafer
fabrication operations in house. Firms that perform both design and wafer fabrication are
referred to as integrated device manufacturers (IDM). As wafer fabrication technology
advanced, the cost of production facilities increased dramatically; the cost of a fabrication
facility has risen from $6 million in 1970 (IC Knowledge 2000) to $4.2 billion in 2009 (Global
Foundries 2009). This sharp increase in cost has made it ever more difficult to stay at the
leading edge of process technology. In the mid 1980s, small semiconductor firms began
producing some of their more advanced designs on the manufacturing lines of larger, more
established semiconductor manufacturers that were better able to bear the capital costs of
maintaining a state-of-the-art fab facility. Many Japanese semiconductor firms had substantial
excess manufacturing capacity during this time period, making such production partnerships
particularly attractive (Hurtarte, Wolsheimer, and Tafoya 2007).
89
Share of actual wafer starts reported in SICAS Semiconductor International Capacity Statistics.
174
These production sharing arrangements led to the creation of a new business model through the
emergence of wafer foundries that manufacture semiconductors designed by other firms. At
first, foundries were used by IDMs as an alternative source of capacity for older process
technologies (Kumar 2008). By the late 1980s a number of new semiconductor firms avoided
wafer fabrication by doing all of their manufacturing through foundries. Semiconductor
companies with little or no in-house wafer manufacturing capability are called “fabless” firms.
In general, fabless firms perform chip design and layout, and use foundries and other contractors
for mask production, wafer fabrication, packaging, and testing. The fabless business model has
grown quickly over the last 30 years, accounting for 24 percent of total semiconductor industry
revenue in 2009, as shown in Figure 3.90 Since the largest foundries are located in Asia, and the
largest fabless semiconductor producers are located in North America and Europe, the growth of
the fabless model has increased the internationalization of semiconductor production.91 Although
the fabless share of the global semiconductor industry only edged up from 2006 to 2008, as new
process technologies continue to raise the costs of fab facilities, the prominence of the fabless
model may well increase even more. Indeed, AMD, the second largest microprocessor producer,
spun off its manufacturing division as an independent foundry company in 2009, boosting the
fabless share of the industry (Clendenin and Yoshida 2002).92
Implications for Price Measurement
The extremely fast pace of technological change in semiconductor manufacturing poses a large
challenge to quality-adjusted price measurement. Aizcorbe (2002) demonstrates the difficulty
government price indexes have had in tracking rapid price declines in finished semiconductors.
However, as just described, technological advance in semiconductor production proceeds in
discrete, measurable steps, in contrast to continuous and difficult to measure quality
improvements seen in other industries (Flamm 1993). This discrete nature of technological
advance in the semiconductor industry makes it possible to control for quality changes, given
detailed enough data on product characteristics. In this study we construct constant-quality price
indexes for wafer fabrication using quarterly pricing data that include the most relevant aspects
of process technology: wafer size, line width, and logic family. We also control for the number
of layers used in constructing the chip, a proxy for design complexity.
This section has also documented the increasing internationalization of the semiconductor supply
chain coinciding with offshoring various steps in the production process and the growth of the
fabless model of semiconductor production. Houseman (2007) describes the challenges faced by
statistical agencies attempting to measure price changes when producers switch suppliers,
particularly when the suppliers are located abroad. In particular, substitution toward low-cost
suppliers is likely to be missed in standard price index calculations (see below for a more
90
Note that the shares in Figure 3 are likely to understate the extent of fabless production activity because
companies must derive 75 percent or more of their semiconductor revenue from fabless production. Many
companies not counted as fabless nevertheless rely heavily on foundries.
91
In 2008, the 5 largest foundries (accounting for 84 percent of foundry revenue) were all located in Asia. Of the 25
largest fabless semiconductor companies (accounting for 75 percent of fabless revenue), 19 were located in North
America or Europe. These figures were calculated from proprietary reports from iSuppli and GSA, respectively.
92
A recent report (IC Insights) predicts that between 2008 and 2013, total foundry sales will grow at double the rate
of the overall semiconductor industry. A recent report (IC Insights) predicts that between 2008 and 2013, total
foundry sales will grow at double the rate of the overall semiconductor industry.
175
detailed discussion), understating the rate of input price decline. As semiconductor production
technology advances and the fabless business model becomes more prominent, it is likely that
these price measurement challenges will remain relevant in the foreseeable future.
In the remainder of this paper, we focus on foundry wafer production, leaving analysis of IDM
production for future work. We make this choice for practical reasons. Our pricing data include
only wafer purchases from foundries, though those purchases could have been made by fabless
firms or IDMs choosing to use foundry suppliers. Also, the issue of within-firm transfer pricing
raises a number of complications that are beyond the scope of this study and makes data
collection essentially impossible.
DATA SOURCES AND DESCRIPTIVE RESULTS
To construct the price indexes used in our analysis, we require information on prices paid and
quantities purchased for foundry services, specified by the characteristics relevant for pricing.
We obtain prices from a survey conducted by the Global Semiconductor Alliance (GSA) and we
calculate quantities by merging several different sources.93 Observations are quarterly, and our
data span the period 2004-2008. Descriptive results demonstrate the importance of controlling
for process technology. They also reveal substantial shifting of production toward lower cost
countries.
Wafer Pricing Survey
Our primary dataset consists of 7,455 individual responses to GSA’s Wafer Fabrication & BackEnd Pricing Survey, provided to us for 2004-2008.94 The survey has been conducted quarterly
since 2004 and provides extensive detail on contracts for foundry services, including key
technological features, foundry location, price paid, and volume for a diverse set of foundry
customers. The survey responses account for a representative sample of about 20 percent of the
wafers processed by the foundry sector.
As shown in Table 1, we drop observations missing key variables. We also drop observations
reporting prices for engineering runs, preliminary fabrication before volume production. To
focus on substitution between onshore and offshore production, and between offshore locations,
we retain only contracts for production at the major offshore locations (Taiwan, Singapore, and
China), U.S. foundry contracts, and European contracts for comparison.95 A small number of
observations with internally inconsistent responses are dropped, as are the handful of
observations on 100mm wafers–a very dated technology. All told, we use 5,464 observations for
index construction.
93
GSA is a semiconductor trade association whose membership includes fabless producers and IDMs. Its survey is
administered to members and nonmembers.
94
Individual respondents are not identified in our data.
95
Significant omissions from the global foundry industry are Japan and Korea. Our approach to estimating capacity,
described below, does not allow us to assign reasonable weights on technologies in Korea. Our preliminary price
index for Japan behaved erratically, and suggested that the product composition was changing in a way not captured
by our data. We have obtained more detailed data extracts that may assist in alleviating this problem in subsequent
versions.
176
Descriptive Price Results
Descriptive statistics for key variables in the resulting dataset are shown in Table 2. We observe
273 prices per quarter, on average. Wafer prices average $1,575 over the period covered.
Interestingly, no substantial time trend is evident before adjusting for composition. The average
contract was for 2,307 wafers, and the average contract size climbs over time. The number of
layers per wafer also rose significantly over the period studied, from 23 in 2004 to 28 in 2008,
reflecting a trend toward foundries handling increasingly complex products.
The changing technological characteristics of the fabrication process are evident in the statistics
for wafer diameter and geometry. Pilot lines for 300 mm wafers were first introduced in 2000
and the share for this emerging technology rises from 3.5 percent of contracts to 20 percent of
contracts over the survey. Similarly, new generations of lithography increase in penetration over
time: 90 nanometer technology reached volume production in the overall semiconductor industry
in 2004 and slowly gained share in the foundry market, ending at 7 percent in 2008; 65
nanometer contracts were just emerging in 2008.96 Meanwhile, older technologies, with
processes above 250 nanometers, dwindle in prominence from 45 percent in 2004 to 28 percent
in 2008. 92 percent of contracts reported in the survey are for CMOS technology, but prices are
available for other processes as well.
A challenge with the GSA pricing survey is sporadic reporting for some technologies in certain
geographic regions, despite independent evidence that such production existed. For such cells
where we believe there was production (based on our capacity database described in the next
subsection) we linearly interpolate prices using values from surrounding periods or extrapolated
based on higher-level prices.97
Quantities and the Shifting Geography of Production
To construct a price index, we need to weight individual price observations by quantity.
Although the GSA survey includes information on the size of each order, some gaps in reporting
remain. This makes weights based on the GSA data unstable at quarterly frequencies. As an
alternative, we construct weights based on global foundry capacity. Although capacity is an
imperfect proxy for actual production or purchases, we must choose between erratic sales
measures and highly credible capacity estimates. Our baseline index uses the latter.
The Gartner Semiconductor Fab Database provided us with quarterly capacity data from 2004 to
2007. For specific fabs, key features are reported, including planned wafer start capacity,
minimum line width, operating status, and whether the fab was operating as a foundry. We
96
2004 and 2007 mark the years when volume production of DRAM began at 90nm and 65nm, respectively
(International Technology Roadmap for Semiconductors 2007.
97
Note that the alternative, dropping these periods for lack of directly observed prices, is not neutral, since it
amounts to 1) assuming the product mix within the industry is different than we know it is, and 2) throwing out price
information from this period for cells with similar technology or geography. See discussion of this approach in
Gordon (2006).
177
extended these data with GSA’s 2009 IC Foundry Almanac, which provides a snapshot of
capacity and technology by fab as of 2009.
Merging these datasets gives us a preliminary set of weights, but we address three remaining
shortcomings. First, Gartner only reports planned capacity by fab and ramp-up status, leaving
the contours of the ramp-up process unknown. Fortunately, many major foundries provide
quarterly information on actual operational capacity, showing the actual path of capacity as
equipment is added incrementally. We employ these directly reported capacities, when
available, and add a comparable ramp-up period to fabs for companies without direct reporting.98
Second, the data do not distinguish CMOS production quantitatively, though GSA does indicate
whether a fab uses CMOS and other processes. Since CMOS prices behave rather differently
than non-CMOS prices, we assigned a weighted average of the CMOS and non-CMOS prices to
each fab for the technology in operation, using overall industry weights from the GSA. Third, in
the Gartner fab database, we only observe the minimum line width in use at a fab, but we know
that fabs often operate multiple geometries one time. This raises the possibility that we
overweight leading edge technologies. On the other hand, it is important to bear in mind that we
only observe capacity, not actual production. Since capacity utilization is higher for leading
edge geometries, the application of capacity weights generates a bias in the opposite direction–
toward underweighting these geometries.99
Table 3 compares two aggregate measures of foundry capacity, constructed as just described, to
industry estimates from other sources. First, wafer fab capacity as reported to the SICAS survey
suggests our wafer fab measure is not fully capturing the overall size of the sector. However, the
growth rate from 2004 to 2008 for the measure constructed from our bottom-up approach is very
close to the SICAS measure, suggesting we are catching the overall trend in industry capacity.
Our measure of revenue is also somewhat lower than the measure of foundry company revenue
published by the consultancy iSuppli. This may simply reflect that not all foundry revenues are
for the services we are studying. Table 4 shows shifting revenue weights among the largest
offshore foundry suppliers. While Taiwan’s share falls somewhat, China and Singapore both
gain revenue share, representing movement toward lower-cost foundry locations.
PRICE INDEX RESULTS
This section presents our price index calculations using the database just described. The level of
detail in our data allows us to adjust for differences in physical product attributes. In addition,
since our data also include foundry location, we are able to isolate the effect of shifting
production across countries on the average wafer price. We find that substitution across countries
may account for no more than a 0.8 percentage point per year decline in the average wafer price.
Our findings also support the established importance of careful quality adjustment to capture the
effects of rapid technological change on semiconductor prices.
98
Ramping new capacity to volume production typically takes 12 months (Semiconductor Industry Association
2007).
99
Utilization on fab lines using 90nm and smaller geometries was 94 percent in 2007, noticeably higher than the 86
percent utilization for larger geometries (SICAS 2008).
178
Fisher Matched Model Index
Our dataset includes price information by detailed semiconductor wafer type and source country
at the quarterly frequency. As discussed in the “Semiconductor Production” section, a wafer’s
process technology (defined by wafer size, line width, and logic family) determines its
performance, along with circuit design. Process technologies proceed in discrete steps, so our
detailed data on prices by process technology yield a time series of price observations for each
wafer type, with attributes held constant over time. This high level of detail allows us to
construct a matched model price index tracking quarterly price changes for each wafer type.
The matched model index is calculated as a Fisher index of price relatives for each process
technology and country pair. First we calculate Laspeyres and Paasche indexes, respectively, as
(1)
P = ∑i ∑ j s
t
L
t −1
ij
pijt
pijt −1
−1
(2)
−1
t
⎤
⎡
⎞
⎛
p
ij
PPt = ⎢∑i ∑ j sijt ⎜ t −1 ⎟ ⎥ ,
⎜p ⎟ ⎥
⎢
⎝ ij ⎠ ⎦
⎣
where i represents process technology, j represents source country, t is time (quarter), and p is
the average price for a given process technology, country, and quarter in the GSA survey.100 s is
the share of total output value in time t accounted for by wafers in the relevant process
technology and country cell, calculated using our capacity database. As the Laspeyres index
overstates price changes and the Paasche understates them, it is advisable to construct the Fischer
index, which is a geometric mean of the Laspeyres and Paasche indexes.
(3)
PFt = PLt ⋅ PPt .
We normalize the index to 100 in the first quarter of 2004.
The procedure just described treats observations from different source countries as separate
“models” by calculating separate price relatives by country. This parallels the treatment of
prices across outlets in the U.S. CPI, and is subject to similar assumptions (Reinsdorf 1993).
When a new process technology and country combination appears, it is assumed that any
difference in the price level across countries for that process technology entirely reflects quality
differences, where “quality” refers to any unmeasured attribute of the wafer or transaction that
makes one production location more attractive than another. This is the “link-to-show-no-pricechange” method in Triplett’s (2006) classification of linking methods for matched model
indexes. This linking strategy is based upon the assumption that the law-of-one-price holds for
quality adjusted units across outlets. As we argue below, there is reason to believe that this
100
Note that we use price per layer for the results presented here to account for the increased cost of producing more
complex wafers containing more layers. As we expect, an index based on price per wafer falls somewhat more
slowly, but the qualitative conclusions using price per wafer are the same as those presented here.
179
assumption does not hold in the semiconductor wafer fabrication industry, potentially leading the
standard matched model index to understate the true rate of price decline.
As expected, entry and exit of products is a prominent feature of the data. As shown in Table 5,
27 cells are new entrants in the 2004–2008 period, and 23 cells are exits. This raises the
challenge of estimating price changes for the first and last periods in the series for a large share
of the data. However, because our data are high frequency (quarterly), the number of entrants or
exits in any given quarter is small, at 2.5 on average. In addition, the weights on these periods
are small as new technologies ramp up gradually.
Table 6 presents our price index calculations. Column (1) contains the Fisher matched model
index just described. We present the quarterly index, yearly averages, and the average yearly
change between 2004 and 2008. The index falls by 12.6 percent per year. As has been known
since at least Flamm (1993), Grimm (1998), and more recently Aizcorbe (2002), quality
adjustment of prices for semiconductors, and indeed for all high-tech products, is critical. In
particular, bear in mind (see Table 2) that the average price change before adjusting for product
composition was slightly positive. The substantial differences across countries points to the
necessity of accurate weights by country.
Relaxing the Location as Quality Assumption
Our previous index maintained the assumption that price differences across countries for
otherwise identical goods reflect unspecified differences in quality. We now make the opposite
assumption: price differences reflect genuine price dispersion across goods of identical quality.
Formally, this means that we calculate unit values by technology, averaging across observations
from different countries. As a result, substitutions toward low-cost producers will be reflected in
the average product price. These two assumptions bracket the truth, which likely lies in between.
We consider this alternative index because the location-as-quality assumption can lead to biased
estimates of price changes under certain circumstances. Consider the convenient example of a
situation in which two countries exhibit similar price trends for a given wafer type, but one has a
consistently lower price level. Under the approach in the previous section, any shift toward the
lower-cost country’s foundries will have no effect on the aggregate price index, since the prices
decline at the same rate in both countries. The linking procedure implicitly assumes that the
savings accrued in shifting supplies are offset by lower quality of the goods being purchased. If,
however, the goods are actually identical, then the shift to the lower-cost country represents a
genuine price drop for the relevant customer. The standard matched model linking approach
misses this price drop achieved in switching suppliers, and thus understates the true rate of price
decline. This is the so-called “outlet substitution bias” discussed in the Boskin Commission
report (Boskin et al. 1996).
To address this, we follow Reinsdorf (1993) and calculate an average price index across
outlets.101 This index is motivated by the opposite quality assumption of the index presented
above. If models are very narrowly defined, one can assume that quality for a given model is
101
Ideally, one would be able to directly observe particular buyers substituting between different outlets. Since our
data do not include purchaser identifiers, directly observing substitution is not possible.
180
identical across outlets. In our context, this amounts to assuming that a given process technology
is identical across foundries in different countries. If this assumption is correct, then there is no
reason to distinguish price relatives by country. Instead, we calculate average prices across
countries for each process technology.
(4)
Pi t = ∑ j wijt pijt ,
where w is country j’s fraction of the total number units of process technology i produced at time
t. We then generate price relatives of these average prices for each process technology and use
them to generate a Fisher price index as described above. This approach is able to capture the
effect of substitution toward low-cost countries as the weights on the lower prices increase with
substitution.
If demand for wafers is shifting toward low-cost suppliers, and the matched model is missing this
substitution effect, we expect to find that the average price index declines more quickly than the
matched model index. The results are presented in Column (7) of Table 6. The index falls by
13.4 percent per year, which is 0.8 percentage points faster than the matched model index in
Column (1). This result supports the notion that outlet substitution bias causes the standard
measure to understate the price declines for wafer fabrication, suggesting an outlet substitution
problem no bigger than 0.8 percentage points per year. Note, however, that the scale of quality
change over time is much larger, as indicated by the sharp overall price declines.
This result should be interpreted with a number of caveats in mind. Both the law-of-one-price
assumption and the alternative assumption of uniform quality across countries are extreme. The
data likely reflect both quality differences across countries and some persistent quality-adjusted
price differences. Thus, the two approaches bound the true quality-adjusted price change, and
the difference between them is an upper bound on the effect of outlet substitution. This
discussion raises the question of why quality-adjusted price differences should be able to occur
in equilibrium. In the semiconductor fabrication market, a number of observations support the
idea that quality-adjusted price differences can persist over time. There have been substantial
shifts toward low-cost countries. This behavior suggests the presence of quality-adjusted
discounts at the low-cost countries. Why might that be? Although Reinsdorf (1993) discusses
the role of costly information gathering in generating real price dispersion, we think that this
explanation is unlikely to hold in a market as concentrated as this one. Rather, we propose an
alternative reason for price dispersion based on the particular characteristics of the wafer
fabrication industry.
Very large fixed costs are incurred when getting a production line up to capacity with a given
design. Discussions with engineers at a large U.S. fabless firm indicate that it takes a large
number of sensitive calibrations to fabricate a particular design on a particular production line.
This creates substantial start-up cost, such that semiconductor firms are very reluctant even to
switch production lines within the same foundry, much less to move a product to a different
foundry. This fact, coupled with the nature of new product introduction across countries leads us
to a potential explanation for equilibrium price dispersion.
181
Consider the price plots presented in Figure 4. The top panel plots prices by country for a
leading edge technology. Taiwan entered the market first, with a high price. Singapore and
China each entered later, each at a lower price level. In spite of the increased competition from
competitors entering the market, the Taiwanese price continued to decline at a steady rate,
maintaining a roughly constant price differential relative to the others. A similar pattern for a
more mature process technology is apparent in the bottom panel of Figure 4, in which a roughly
constant price differential is maintained between the U.S. and Taiwan relative to Singapore and
China.
To understand the implications of these observations, consider only Taiwanese and Chinese
foundries for simplicity. If a given design requires the newest technology, it will have to be
produced in Taiwan. In two years’ time, when the Chinese foundry brings the same process
technology on line, they charge a lower price in order to win market share away from their
Taiwanese competitors. However, the lower wafer price in China does not outweigh the fixed
cost of moving the existing products from Taiwan. The Taiwanese foundry can maintain a
discretely higher price without losing its existing business, and only new products using the now
year-old technology will go to the lower priced Chinese foundry. The Chinese foundry may
adopt the new technology more slowly due to a relative lack of technical expertise or due to U.S.
export license restraints on advanced semiconductor fabrication equipment going to China
(Electrical Engineering Times 1998). In any case, the presence of large fixed costs of switching
foundries coupled with staggered entry into a given technology makes persistent quality-adjusted
price differences across countries possible.
Hedonic Price Index
To check the robustness of our results, we next generate a hedonic price index. Table 7 presents
some information on the importance of the characteristics we observe. We regress log price per
wafer on indicators for foundry location, technological characteristics, contract size, and quarter
indicators using the 5,000 observations on contracts for CMOS technology.102 All of these
variables have a noticeable effect on prices and are estimated precisely. Collectively, they
account for 88 percent of the variation in wafer prices.
The point estimates on foundry location and process technology appear to be reasonable.
Controlling for technology, China has markedly lower prices than Taiwan, which serves as the
baseline case in the regression. Singapore’s prices are moderately lower than Taiwan’s, while
U.S. and European prices are substantially higher. Production using more advanced technologies
clearly commands a higher price. Compared to the baseline case of production on 200 mm
wafers with 180 nm geometry, production on larger (300 mm) wafers and production with
narrower line widths is significantly more expensive. More overall layers per chip, and more
metal layers in particular, both proxies for the complexity of the circuitry, also drive up the price.
Finally, contracts involving a greater scale of production do appear to draw a volume discount;
102
As mentioned above, non-CMOS technology is generally used in specialized niche markets. Although we do use
non-CMOS prices when calculating industry price indexes, we omit them here for simplicity of exposition. Results
for non-CMOS prices, not shown, indicate that location explains little of the variation in pricing, but technological
characteristics do play a role.
182
other things equal, doubling contract size would be expected to reduce wafer costs by 5.5
percent.
Like the matched-model index, the hedonic index also falls rapidly, though the 11 percent
average yearly rate of decline is 2 percentage points short of the rate for the matched model.103
From this we conclude that our baseline results are fairly robust to choice of price index
construction methodology. The hedonic specification also controls for characteristics not
addressed in the matched model index, which suggest that contract size and the composition of
layers contracted does affect pricing. The regression statistics indicate that these features explain
over 80 percent of the variation in prices.
Official Indexes
For completeness, this section compares our results to the Bureau of Labor Statistics’ (BLS)
price series for imported semiconductors. The BLS’ International Price Program (IPP) publishes
a price index for Harmonized System code 8542, Electronic integrated circuits. These include
microprocessors and memory, the final products of the semiconductor production chain.
IPP draws its sample from Customs lists at the more detailed 10-digit Harmonized System
level.104 For instance, until recently, IPP would draw a sample of establishments whose
product(s) are recorded under the just phased-out HS classification 8542.21.80.05 for
“unmounted chips, die, and wafers.” Price indexes are calculated at this more disaggregated level
and IPP then aggregates across the price relatives to produce the published index. Unfortunately,
this more detailed data is sealed to outside researchers for confidentiality reasons.
Perhaps the measurement challenge for IPP is to control for quality improvements in ICs. We do
this via a matched model price index that controls for several important performance-related
characteristics of wafers. IPP does not necessarily observe as many characteristics of each IC,
but it does have a potentially promising way to identify quality improvements. At least some
respondents provide BLS staff with their own internal product code assigned to the surveyed
item. It is likely that new, higher quality products would receive a new product code. If IPP
observes that the product code attached to the surveyed item changes, it will follow up with the
respondent to ask what the price of the new product would have been last month so that it can
record the true price change for the quality-enhanced good. These follow-ups based on observed
changes in firm product codes appear to be one of the principal ways by which IPP adjusts
goods, at least in HS 8542, for quality improvements.105
The ICs observed by IPP are not directly comparable to the wafers studied in this paper. To see
this more clearly, it is useful to recall that we can break up the production of ICs into four
stages—design, wafer fabrication, test, and assembly. Our data pertain to the input produced in
103
Aizcorbe et al. (2003) find a similar result for microprocessors.
This discussion draws on a number of conversations with Sonya Wahi-Miller of the IPP. We are very grateful for
the time she spent educating us on the IPP’s procedures. Any errors in our characterization of the IPP, however, are
our own.
105
Thus far, we have been unable to obtain information on how often this procedure is generally used in generating
the HS 8542 index.
104
183
stage two whereas IPP measures the price of final output shipped at the conclusion of stage four.
Nonetheless, it is instructive to ask how average price per wafer compares to the IPP estimate of
the price of the finished product.
Table 6 Column (9) presents the IPP index by quarter over the period 2004-2008. Over this time
period, the index falls on average 2.9 percent per year. Even though this is not directly
comparable to our indexes, the discrepancy is quite large. It would imply that the prices in the
remainder of the production chain (development, wafer test, and assembly) fall implausibly
slowly. Consider, for instance, that recent research has found price declines that approach 40–50
percent per year for finished semiconductors sold in the United States (see, among others,
Aizcorbe [2002, Table 1]). This work suggests that prices at other stages of the production chain,
such as test and assembly, actually fall faster than the price of wafer fabrication, which contrasts
starkly with the message sent by the IPP series. A critical task for future work is to dig deeper
into the sources of these discrepancies. In particular, it seems worthwhile to investigate whether
the IPP’s follow-up procedure for product code changes does in fact effectively capture key
quality improvements.
CONCLUSION
Our analysis exploits a rich new dataset to calculate constant quality price indexes for processed
semiconductor wafers. We calculate a matched model price index, finding that wafer prices fall
on average by 12.6 percent per year. Given that average prices, unadjusted for quality, remain
fairly constant over the time period, the sharp yearly price decline demonstrates the importance
of careful quality adjustment in this industry. Our results support the conclusion of numerous
previous studies that official statistics substantially understate the rate of semiconductor price
decline.
Since our dataset includes information on the source country for wafer purchases, we can also
measure how geographic changes in sourcing patterns affect price measurement. Our approach
is analogous to Reinsdorf’s (1993) measurement of retail outlet substitution bias in the CPI. We
calculate an average price index that captures the effects of shifting sourcing patterns toward
wafer foundries in low-cost countries. Our results imply that the baseline matched model
approach understates the yearly price decline by at most 0.8 percentage points.
Although this problem is not overwhelming, particularly in comparison to the much larger issue
of quality adjustment in the semiconductor industry, it is suggestive that continued shifts in
international sourcing patterns will cause the problem to persist and potentially grow. Our
findings here should motivate research into other industries that have seen large shifts in
sourcing patterns across countries. Since there are large fixed costs of shifting suppliers in
semiconductor production, the finding here may be smaller than the bias in more footloose
industries that can substitute quickly in response to smaller price differences. Note however, that
future analyses will need to motivate the assumption of persistent quality adjusted price
differences across suppliers, as we do here.
184
Figure 1: Moore’s Law – Intel Processors
Sources: http://www.intel.com/technology/timeline.pdf
http://www.intel.com/pressroom/kits/quickreffam.htm
Figure 2: Technology Cycle – TSMC Sales by line width
80%
70%
500nm
TSMC Fraction of Total Sales
60%
50%
350nm
130nm
250nm
40%
150nm
90nm
180nm
65nm
30%
20%
10%
0%
1997
1998
1999
2000
2001
2002
2003
Source: TSMC quarterly reports
185
2004
2005
2006
2007
2008
2009
Figure 3: Growth of the Fabless Business Model
Fabless Share of Total Semiconductor Revenue
20%
15%
10%
5%
0%
1986
1988
1990
1992
1994
1996 1998
2000
2002
2004
2006
2008
Sources: Global Semiconductor Association (GSA) and Semiconductor Industry Association (SIA)
186
Figure 4: Price Differences Across Locations
Leading Edge Technology - CMOS, Wafer Size: 300mm, Line Width: 130nm
6000
5500
United States
Price Per Wafer ($)
5000
4500
4000
3500
Taiwan
3000
2500
Singapore
2000
China
1500
2004
2005
2006
2007
2008
Mature Technology - CMOS, Wafer Size: 200mm, Line Width: 180nm
1800
Price Per Wafer ($)
1600
1400
United States
1200
1000
Taiwan
800
China
Singapore
600
2004
2005
2006
187
2007
2008
Table 1: Dropped Observations
Total observations
7455
Used in analysis
5464
Dropped
Missing:
foundry location
wafers purchased
price
Other reason:
engineering run
location
100mm wafer
inconsistent
1991
813
19
19
778
499
3
3
Note: there may be multiple reasons to drop a particular observation
Table 2: Descriptive Statistics
Mean
Price Per Wafer ($)
2004
Std. Dev
Yearly Means
2005
2006
2007
2008
1575.40
1145.54
1,576.58
1,609.53
1,502.86
1,545.03
1,655.18
Number of Wafers Contracted
2307
7514
1924
2357
1941
2710
2627
Number of Layers Per Wafer
Metal Layers
25.74
4.77
7.57
1.81
23.25
4.23
24.64
4.55
25.79
4.75
26.64
4.97
27.93
5.27
Wafer Size
150 mm or less
200 mm
300 mm
0.14
0.76
0.10
0.35
0.42
0.30
0.17
0.80
0.03
0.17
0.77
0.06
0.15
0.79
0.06
0.12
0.76
0.12
0.10
0.70
0.20
Line Width
65 nm
90 nm
130 nm
180 nm
250 nm
older vintage
0.00
0.03
0.23
0.25
0.13
0.36
0.06
0.16
0.42
0.43
0.34
0.48
0.00
0.00
0.14
0.26
0.13
0.45
0.00
0.08
0.18
0.27
0.16
0.38
0.00
0.01
0.22
0.26
0.12
0.38
0.00
0.03
0.27
0.26
0.12
0.31
0.01
0.07
0.32
0.22
0.09
0.28
CMOS process
0.92
0.28
0.92
0.92
0.92
0.91
0.91
5464 Observations
Source: Authors' calculations based on GSA Wafer Fabrication & Back-End Pricing Survey
188
Table 3: Coverage of Constructed Capacity and Revenue
2004
2005
2006
2007
2008
Wafer Start Capacity
(1,000 Wafers per Week)
SICAS
Constructed
194
123
252
139
285
151
288
172
297
188
Revenue
(US $ Billion)
iSuppli
Constructed
16.6
9.1
16.3
9.0
19.5
9.6
19.7
9.8
20.1
9.9
Source: SICAS, iSuppli, and author's calculations from sources described in text
Table 4: Foundry Revenue and Share for Major Offshore Locations
2004
2005
2006
2007
2008
Revenue ($million)
7232
8517
8549
8668
8432
Taiwan
66.0%
61.7%
62.0%
60.3%
59.8%
China
19.7%
20.4%
20.1%
21.6%
21.7%
Singapore
14.3%
17.8%
17.9%
18.1%
18.5%
Note: Includes pure-play foundries only.
Source: Authors' calculations based on data from GSA, Gartner, and
company reports
Table 5: Entry and Exit Statistics, CMOS Process
country technology cells with data
ave. no. quarterly prices per cell
new entrants
exits
cells with entry or exit
ave. quarters with missing prices
189
74
10.18
27
23
38
5.375
Table 6: Price Index Results
(1)
(2)
Quarter
2004Q1
2004Q2
2004Q3
2004Q4
2005Q1
2005Q2
2005Q3
2005Q4
2006Q1
2006Q2
2006Q3
2006Q4
2007Q1
2007Q2
2007Q3
2007Q4
2008Q1
2008Q2
2008Q3
2008Q4
Overall
100.0
101.5
99.6
93.5
91.3
83.5
81.7
82.0
76.4
74.4
72.4
69.6
70.3
67.9
62.8
59.6
60.4
57.1
58.2
54.2
Taiwan
100.0
101.7
103.2
95.1
86.9
76.9
79.5
79.2
73.6
71.6
69.4
65.9
67.1
63.3
58.7
55.5
55.7
51.9
52.5
49.2
Year
2004
2005
2006
2007
2008
98.6
84.6
73.2
65.2
57.5
100.0
80.6
70.1
61.2
52.3
93.6
93.0
79.5
71.9
65.4
-12.6%
-14.9%
-8.6%
Avg. Yearly
Change '04-'08
(3)
(4)
(5)
(6)
(7)
(8)
(9)
USA
100.0
97.5
97.2
95.5
93.0
95.2
80.0
79.3
69.4
70.8
66.4
64.1
65.6
59.1
56.0
52.2
58.2
57.1
69.2
63.4
Europe
100.0
108.5
94.6
89.1
89.5
85.1
86.8
92.1
87.2
82.0
82.0
82.1
87.6
90.0
88.4
84.1
83.7
83.3
85.3
82.9
Average
Price Index
100.0
100.7
98.4
89.5
87.7
79.1
79.5
77.8
72.8
70.4
68.7
66.3
66.9
64.8
59.7
56.5
57.3
54.3
55.3
51.4
Hedonic
Index
100.0
99.5
97.7
91.3
87.4
87.3
83.8
82.7
78.4
74.1
73.6
71.0
69.2
67.6
65.3
64.5
65.0
61.9
59.6
59.7
BLS IPP
HS 8542
100.0
98.3
97.1
95.9
95.5
95.1
93.9
93.5
94.0
93.8
94.6
95.3
93.3
88.8
90.0
90.3
88.5
87.5
85.8
85.6
101.6
92.5
80.9
74.5
66.2
97.6
86.9
67.7
58.2
62.0
98.1
88.4
83.3
87.5
83.8
97.2
81.0
69.5
62.0
54.6
97.1
85.3
74.3
66.6
61.6
97.8
94.5
94.4
90.6
86.9
-10.2%
-10.7%
-3.9%
-13.4%
-10.8%
-2.9%
Fisher Matched-Model Indexes
China
Singapore
100.0
100.0
99.9
102.7
90.1
101.7
84.3
102.1
100.5
101.5
95.2
94.1
85.5
88.7
90.6
85.8
83.5
82.0
77.8
84.2
78.0
80.8
78.6
76.4
77.7
75.7
77.4
77.1
67.2
74.8
65.3
70.1
66.3
71.7
63.9
68.2
68.2
65.0
63.1
59.7
190
Table 7: Descriptive Wafer Price Regression Results
dependent variable: log of price per wafer
Variable
Coefficient
Std. Err.
t-Stat
Foundry Location
China
United States
Europe
Singapore
-0.272
0.218
0.119
-0.062
0.019
0.014
0.018
0.012
-14.59
15.58
6.54
-5.30
Wafer Size
150 mm
300 mm
-0.344
0.645
0.015
0.014
-22.20
47.68
Line Width
≥ 1000 nm
800 nm
600 nm
450 nm
350 nm
250 nm
130 nm
90 nm
65 nm
-0.696
-0.353
-0.358
-0.355
-0.194
-0.092
0.306
0.511
0.737
0.038
0.027
0.022
0.019
0.013
0.012
0.012
0.025
0.050
-18.24
-13.31
-16.12
-18.35
-14.59
-7.74
26.31
20.19
14.63
layers per wafer
no. metal layers
log wafers contracted
0.012
0.057
-0.055
0.001
0.004
0.002
13.83
14.30
-32.92
6.743
0.030
223.47
constant
R-squared
Observations
0.8773
5000
Specification also includes quarterly indicator variables
non-CMOS production not included
Baseline case (omitted category) is Taiwan, 200mm, 180nm
191
References
Aizcorbe, Ana. 2002. “Price Measures for Semiconductor Devices.” Research Paper Series No.
2002-13. Washington, DC: Board of Governors of the Federal Reserve System
Aizcorbe, Ana, Carol Corrado, and Mark Doms. 2003. “When Do Matched-Model and Hedonic
Techniques Yield Similar Price Measures?” Working Paper No. 2003-14. San Francisco:
Federal Reserve Bank of San Francisco.
Boskin, Michael J., Ellen R. Dulberger, Robert J. Gordon, Zvi Griliches, and Dale Jorgenson.
1996. Toward a More Accurate Measure of the Cost of Living. Final Report to the Senate
Finance Committee from the Advisory Commission to Study the Consumer Price Index (the
Boskin Commission). Washington, DC, December 4.
Brown, Clair, and Greg Linden. 2006.“Offshoring in the Semiconductor Industry: A Historical
Perspective.” In Brookings Trade Forum, 2005, Lael Brainard and Susan M. Collins, eds.
Washington, DC: Brookings Institution Press, 279–333.
Clendenin, Mike, and Junko Yoshida. 2002. “Game, Set, Match for Foundries? Even Top-Tier
IC Makers Partnering as CMOS Process Rigors Intensify.” Electronic Engineering Times
March 11: 1–2.
Diewert, W. Erwin. 1998. “Index Number Issues in the Consumer Price Index.” Journal of
Economic Perspectives 12(1): 47–58.
Electronic Engineering Times. 1998. “U.S., China at Odds on Fab-Gear Export.” 1004: 8.
Flamm, Kenneth. 1993. “Measurement of DRAM Prices: Technology and Market Structure.” In
Price Measurements and Their Uses, Vol. 57 of National Bureau of Economic Research
Studies in Income and Wealth, Murray F. Foss, Marilyn E. Manser, and Allan H. Young,
eds. Chicago and London: University of Chicago Press.
Global Foundries. 2009. “Fact Sheet: Fab 2 Module 1.”
http://globalfoundries.com/newsroom/2009/Fab_2_fact_sheet.pdf (accessed March 2009).
Gordon, Robert J. 2006. “The Boskin Commission Report: A Retrospective One Decade Later.”
International Productivity Monitor 12: 7–22.
Grimm, Bruce T. 1998. “Price Indexes for Selected Semiconductors 1974-96.” Survey of Current
Business 78(2): 8–24.
Houseman, Susan. 2007. “Outsourcing, Offshoring and Productivity Measurement in United
States Manufacturing.” International Labour Review 146(1-2): 61–80.
Hurtarte, Jeorge S., Evert A. Wolsheimer, and Lisa M. Tafoya. 2007. Understanding Fabless IC
Technology. Oxford: Elsevier.
193
IC Knowledge. 2000. “Can the Semiconductor Industry Afford the Cost of New Fabs?”
http://www. icknowledge.com/economics/fab_costs.html (accessed March 2010).
Kumar, Rakesh. 2007. “The Business of Scaling.” IEEE Solid-State Circuit Society News 12(1):
22–27.
———. 2008. Fabless Semiconductor Implementation. New York: McGraw-Hill Professional.
Reinsdorf, Marshall. 1993. “The Effect of Outlet Price Differentials on the U.S. Consumer Price
Index.” In Price Measurements and Their Uses, Vol. 57 of National Bureau of Economic
Research Studies in Income and Wealth, Murray F. Foss, Marilyn E. Manser, and Allan H.
Young, eds. Chicago and London: University of Chicago Press.
Scott, Allan J., and David. P. Angel. 1988. “The Global Assembly-Operations of US
Semiconductor Firms: A Geographical Analysis.” Environment and Planning A 20(8):
1047–1067.
Semiconductor Industry Association. 2007. “International Technology Roadmap for
Semiconductors.” International SEMATECH, Austin, TX.
Semiconductor International Capacity Statistics (SICAS). 2008. Statistics Report: Semiconductor
Wafer-Fab Capacity and Utilization. Vessem, The Netherlands. `
Triplett, Jack. 2006. Handbook on Hedonic Indexes and Quality Adjustments in Price Indexes:
Special Application to Information Technology Products. Paris and Washington, D.C.:
Organization for Economic Cooperation and Development.
Turley, James L. 2003. The Essential Guide to Semiconductors. New Jersey: Prentice Hall.
194
Globalization, Prices of IT Software & Services:
Measurement Issues
Catherine L. Mann
Professor, International Business School, Brandeis University
Senior Fellow, Peterson Institute for International Economics
Conference on Measurement Issues Arising from the Growth of Globalization
November 2009
195
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
Imports of Intermediate Parts in the Auto Industry
—A Case Study
Paper written for the “Measurement Issues Arising from the Growth of Globalization”
conference, November 6–7, 2009, in Washington, D.C.
Thomas H. Klier
Federal Reserve Bank of Chicago
[email protected]
James M. Rubenstein
Miami University (Ohio)
[email protected]
October 23, 2009
Abstract
Intermediate parts contribute roughly 70 percent of the value added in production of motor
vehicles. Carmakers like Ford and General Motors once made many of these parts in house, but
now procure most of them from independent producers. Outsourcing of parts extends globally,
with more than one-fourth of the parts in vehicles assembled in the United States imported from
other countries. This paper describes the growing role of motor vehicle parts imports for U.S.based light vehicle assembly. Imports of motor vehicle parts have increased to both substitute for
U.S.-based parts production as well as to complement U.S.-based vehicle production of foreign
producers. The paper assesses the effect of imports on production costs by distinguishing highand low-cost source countries. This analysis is supplemented with anecdotal evidence on two key
measurement issues: 1) the globalization of supply chains, as well as the relocation of production
from a developed to a less-developed country; and 2) attendant changes in the structure of
production costs.
The authors thank Cole Bolton, Taft Foster, and Justin Hess for excellent research assistance.
Thanks to Susan Houseman and Richard Lilley for helpful comments.
217
MOTIVATION
The production of an automobile is complex, involving thousands of parts and hundreds of
different companies.106 As many of the intermediate parts cross international borders (some
multiple times, especially between Michigan and Ontario along the U.S. Canadian border),
automobile production is of interest from the vantage point of properly measuring the extent of
offshoring as well as the price of attendant imported intermediate goods.
Intermediate parts contribute roughly 70 percent of the value added in production of motor
vehicles. Carmakers like Ford and General Motors once made many of these parts in house, but
now procure most of them from independent producers. Outsourcing of parts has also been
globalized, with more than one-fourth of the parts in vehicles assembled in the United States
imported from other countries. In the same vein, foreign-headquartered motor vehicle parts
producers have established significant production operations in the United States.
This paper describes the growing role of motor vehicle parts imports for U.S.-based light vehicle
assembly. Imports of motor vehicle parts have increased to both substitute for U.S.-based parts
production as well as to complement U.S.-based vehicle production of foreign producers. The
paper serves as a case study in the context of this conference by focusing on two key
measurement issues: 1) the globalization of supply chains, as well as the relocation of production
from a developed to a less-developed country; and 2) attendant changes in the structure of
production costs. Both examples will be discussed in the context of Mexico as a production
location for motor vehicle parts.
The paper is structured as follows. The following section summarizes the relevant literature. We
then discuss the source for our data on the auto supplier industry. The next section presents the
trends in imports of motor vehicle parts to the United States. We then focus on Mexico as a
source for U.S. motor vehicle parts imports and present two examples illustrating in some detail
the challenges for proper measurement of the imports of intermediate inputs in the auto industry:
the shift of production for a specific product, aluminum wheels, from the United States to
Mexico, and the globalization of the supply chain for an intermediate part, the seat.
LITERATURE
The auto industry is often highlighted as an example of a global manufacturing industry (see, for
example, work at the International Motor Vehicle Program, as well as Sturgeon, Van
Biesebroeck, and Gereffi [2007]). In North America, automobile and parts production has long
been integrated across the U.S. and Canadian border (Weintraub and Sands 1998). Mexico
became an important location for parts production starting in the late 1970s. Montout,
Mucchielli, and Zignago (2007) suggest that the degree of intraindustry trade in the North
American automobile industry increased at the beginning of the 1990s.
106
In this paper the term automobile is used synonymously with “light vehicle,” which is a term frequently used to
summarize vehicles consumers tend to buy. Light vehicles are cars and light trucks, such as pick-ups, SUVs, and
minivans.summarize vehicles consumers tend to buy. Light vehicles are cars and light trucks, such as pick-ups,
SUVs, and minivans.
219
Most of the literature on the North American auto industry describes the evolution of especially
Mexico as an important source of intermediate good production (see, for example, Gilmer and
Canas [2008], Carillo and Contreras [2007], and earlier, Herzenberg [1991] and U.S. Congress,
[1992]). In the last few years there have been a number of papers estimating the share of vertical
and horizontal intraindustry trade in the auto industry (see Montout, Mucchielli, and Zignago
[2007] and Ito and Umemoto [2004]). Regarding the focus of this conference, there is very little
information on the relative production costs of auto parts, including comparisons between
production in the United States and Mexico. The few published examples tend to be dated and
apply to large and complex components, such as engines (see U.S. Congress [1992]).107 Klier and
Rubenstein (2008) provide a recent and comprehensive analysis of the trade flows in auto parts
to and from the United States.
DATA
Detailed Harmonized Tariff System (HTS) code data are available from the U.S. International
Trade Commission. That coding system was enabled by the 1988 Trade Act. It created the HTS
system which authorized 8- and 10-digit codes for imports. This very detailed data source forms
the basis for our analysis of changes in the nature and source of motor vehicle part imports to the
United States. Many parts for motor vehicles are included in HTS code chapter 87, yet they are
scattered throughout a number of other chapters as well.
To generate a comprehensive list of motor vehicle parts based on HTS codes, we painstakingly
combed through all relevant HTS chapters. Our goal was to identify parts intended for use in the
assembly of new light vehicles (so-called light vehicle OEM parts). Therefore we excluded parts
for use in motorcycles, buses, or commercial trucks whenever possible. Despite the incredible
detail available in the 10-digit HTS-code system, there is one major drawback to the data
classification: trade data, just like census-based data on U.S. production, cannot identify where
the parts will be used. Ideally we would like to focus in our analysis exclusively on parts that are
intended for the assembly of new vehicles as opposed to “aftermarket” parts, which are parts that
end up in the retail or wholesale channel (for example, for installation at a car repair shop). For
large and complex parts, such as engines and transmissions, the HTS codes distinguish new from
“remanufactured” parts. While that distinction does not substitute as an identification of OEM
and aftermarket parts—e.g., a new engine can be purchased through a parts dealer—we excluded
all remanufactured parts, as those are not intended for use in the assembly of a new vehicle. Our
list of motor vehicle parts consists of just over 200 individual 8- and 10-digit HTS codes,
representing 10 different 2-digit chapters.
We supplemented the trade data with a plant-level database that describes the geography of
motor vehicle parts production by part in North America. The plant-level database covers 3,179
parts plants in the United States, 416 in Canada, and 673 in Mexico. It represents information
107
The only exception we came across is a comparison of production costs of wiring harnesses for a U.S. and
Mexico location (U.S. Congress 1992, p. 147). Assembly costs of wiring harnesses, a very labor-intensive product,
in the United States around 1990 ranged from $12 to $23. Assembly cost in Mexico varied between $1 and $2;
shipping and inventory added $7.50. All costs are expressed in U.S. dollars.
220
from late 2006 to early 2007 and is the basis for our book on the North American auto supplier
industry.108
TRENDS IN MOTOR VEHICLE PARTS IMPORTS
Figure 1 shows that the value of imports of motor vehicle parts imports (as defined above) more
than doubled between 1996 and 2008. Yet the volume of U.S. light vehicle production between
1996 and 2006 fluctuated in a rather narrow band, between 10 and 12 million units, before
steadily declining to below 8 million units by the end of 2008. As a share of the material costs of
light vehicle assembly (data available from the census of manufactures), imports have increased
noticeably from 29 percent in 1997 to 36 percent in 2002.
Figure 2 breaks out data on U.S. imports of motor vehicle parts by countries of origin. It
identifies the five largest countries in year 2008. The remainder is aggregated into the “rest of the
world” category. There has been a fair amount of movement among the largest source countries
during the last decade and a half. Canada and Mexico, the two NAFTA partners, have
traditionally represented the origin for more than half of all U.S. imports of motor vehicle parts.
In 1996 the two counties represented nearly 60 percent of all U.S. parts imports, with Canada
firmly holding the lead. By 2008 Canada was essentially tied with Japan for rank three among
import source countries, having lost more than 10 percentage points in almost a decade and a
half. Mexico’s share of U.S. motor vehicle parts imports held steady at just below 30 percent; by
2004 it had taken over as the largest source of imports from Canada. China represents the fastest
growing origin of motor vehicle parts imports. It had eclipsed Germany for rank four by 2006
and represented 10 percent of U.S. imports in 2008.
Figure 3 breaks out all motor vehicle parts imports by high- and low-wage countries.109 It
demonstrates the steady growth of imports in motor vehicle parts from low-wage countries
during the last decade and a half. Low-wage countries added about 25 percentage points of
import share during that time. By 2007 the majority of all parts imports originated in low-wage
countries. Stated differently, 69 percent of the growth in motor vehicle parts imports between
1996 and 2008 had originated in low-wage countries. Figure 4 identifies the three largest source
countries for both high- and low-wage countries. Among high-wage countries, Canada’s role has
been shrinking, whereas China has been growing among low-wage countries.
FOCUS: MEXICO
We now look toward Mexico, the largest source of low-wage country imports of motor vehicle
parts to the United States. After a brief recap of the history of the Mexican motor vehicle parts
industry, we illustrate with two specific examples the growth in imports of motor vehicle parts:
the shift of production from the United States to low-wage countries, and the complexity of the
supply chain for intermediate parts.
108
109
See Klier and Rubenstein (2008, pp. 10–13) for a detailed description of the construction of the plant-level data.
High-wage countries consist of Canada, Japan, and all of Western Europe.
221
Maquiladora Plants
The leading suppliers of motor vehicle parts from Mexico have been foreign-owned maquiladora
plants.110 Mexico’s Border Industrialization Program, established in 1965, permitted foreign
companies to import materials from the United States, assemble them in so-called maquiladora
plants, and export them back to the United States without having to pay duty on the raw
materials brought into Mexico, the equipment in the maquiladora plants, or the subassemblies
shipped back to the United States.
U.S. auto parts makers started taking advantage of the maquiladora laws in the late 1970s. GM’s
Packard Electric Division, now part of Delphi, established Conductores y Componentes
Electricos to make wire harnesses, a very labor-intensive part, in Ciudad Juarez in 1978.
Electrical components dominated Mexican early maquiladora production, accounting for twice
as many imports as all other systems combined into the 1990s.
GM’s Inland Division, now also part of Delphi, arrived in Ciudad Juarez in 1978 to make seat
covers and interior trim. Production of seat components expanded rapidly into the twenty-first
century as the three large assemblers of complete seats, Lear, JCI, and Magna, relocated
production of some individual components to Mexico and purchased more individual seat parts
from Mexico-based lower-tier suppliers.
In terms of geography, maquiladora plants are strung out in Mexican cities along the U.S. border,
especially (from east to west) in Matamoros (across the border from Brownsville, Texas),
Reynosa (across from McAllen), Nuevo Laredo (across from Laredo), Ciudad Juarez (across
from El Paso), and Tijuana (across from San Diego). The more easterly cities have attracted most
of the auto parts maquiladoras because of their relative proximity to auto alley. Auto-related
maquiladora production is also clustered in larger northern Mexican cities 100 miles or so south
of the border, such as Nuevo Leon, Monterrey, Chihuahua, and Hermosillo.
According to the Mexico Maquila Information Center, 24 of the 100 largest maquiladoras in
2006 were motor vehicle parts suppliers. The three largest maquiladoras on the list were Delphi,
Lear, and Yazaki, all motor vehicle parts producers. The 24 auto-related maquiladoras together
employed 216,696 workers in Mexico in 2006, including 66,000 at Delphi, 34,000 at Lear, and
33,400 at Yazaki.
Figure 5 demonstrates the changing composition of motor vehicle parts imports from Mexico
during the last decade and a half. The figure is based on HTS-code import data. We aggregated
the individual parts into seven distinct subsystems, such as electrical and drivetrain (engine and
transmission). The data illustrate the large share of electrical parts, even in 2008. To this day
Mexico continues to be by far the largest source of automobile wiring harnesses imported into
the United States (47 percent of all imported wiring harnesses came from Mexico in 2008, down
slightly from 54 percent in 1996). Yet there is evidence that the composition of the types of auto
parts produced in Mexico for U.S. consumption is changing. See, for example, Carillo and
Contreras (2007), who point out that companies such as Delphi have upgraded their production
operations in Mexico “from simple assembly to centralized coordination of functions including
110
This section draws heavily on Klier and Rubenstein (2008, pp. 318–320).
222
sophisticated product design, development, and research (p. 2).” As part of that transformation
Delphi transferred a technical center and its research, design, and product development functions
from Anderson, Indiana, to Ciudad Juarez, Mexico, in 1995 (p. 4).
Shift of Production from the U.S. to Mexico: Aluminum Wheels
We chose the aluminum wheel as an example of a component for which most production has
relocated during the past decade from the United States to low-wage countries. The wheel
represents a rather well-defined stand-alone part. Its production is quite simple. About 70 percent
of production costs are represented by alumina, the raw material, and the processing of it, such as
casting, heat-treating, machining, and painting.111 The aluminum wheel has its own 10-digit HTS
code, although as noted already we are not able to distinguish between OEM and aftermarket
imports. We were also able to obtain some detailed information about the cost structure of
aluminum wheel production in the United States and Mexico from Richard M. Lilley of Lilley
Associates, Inc., which publishes a biannual NAFTA Light Vehicle Road Wheel Survey.
Aluminum is the main material for the construction of wheels, representing roughly two-thirds of
the OEM market for wheels in the world and in North America. Although more expensive than
steel, aluminum has replaced steel during the past quarter-century as the metal of choice for
casting wheels, because it is much lighter and can be more easily shaped into designs that
carmakers prefer.
The mass-produced aluminum wheel is a commodity that is sourced by carmakers on the basis of
price. Carmakers know to the fraction of a penny the cost of each component in the price of a
wheel, especially standard wheels produced in high volumes. Profit margins on high-volume
wheels are extremely small, according to Lilley Associates. Wheel suppliers make a profit
primarily by producing low-volume niche and specialty wheels.
The two leading U.S.-based suppliers of wheels are Hayes-Lemmerz and Superior Industries.
Hayes, a venerable supplier founded in 1908, filed for Chapter 11 in May 2009. Hayes has been
shedding other parts units to focus on wheels. Once the dominant supplier of steel wheels, the
company was slow getting into aluminum. Superior Industries, the other leading U.S.headquartered producer, is the “upstart,” having been founded in 1973 in California. Superior
specializes in the production of aluminum wheels. The company has about one-third of the U.S.
market for aluminum wheels. Like Hayes, Superior has addressed the difficult economic climate
by giving up production of parts other than wheels.
Superior produces aluminum wheels at two factories in the United States, both in Arkansas, and
three in Mexico, all in Chihuahua. Between 2007 and 2009, the company closed three of its
aluminum wheel plants in the United States, one each in California, Kansas, and Tennessee. As a
result, two-thirds of Superior’s North American OEM wheel production is in Mexico and onethird in the United States, according to its 2008 annual report.
111
Once melted, the aluminum is fed to the wheel casting machines. After the aluminum has solidified, the blank
casting is sent to trimming machines and to the heat treatment station. Each wheel is “baked” for several hours to
give it the proper metallurgical structure and strength. Subsequently the wheel is sent to the machining stations for
drilling. Afterward wheels are cleaned and painted.
223
All of Hayes’ North American aluminum wheels are produced in Mexico. Hayes closed its only
U.S. aluminum wheel factory in Gainesville, Georgia, in 2008. The company started wheel
production in Mexico as the minority partner in a joint venture. It has since taken over full
control over that operation.
Other North American producers of aluminum wheels include Alcoa Wheel Products, a division
of Alcoa, which invented the forged aluminum wheel in 1948. It produces aluminum wheels in
Cleveland, Ohio, and Lebanon, Virginia, but most of these wheels are destined for commercial
vehicles and the aftermarket. The company closed an aluminum wheel plant in Beloit,
Wisconsin, in 2008. Several Japanese-owned aluminum wheel suppliers have U.S. operations.
Central Motor Wheels of America and Canadian Autoparts Toyota, both Toyota captives,
produce aluminum wheels in Paris, Kentucky, and, Delta, British Columbia, respectively. These
two companies are the principal supplier of aluminum wheels to Toyota’s North American
assembly operations.112 AAP St. Mary’s, a subsidiary of Hitachi, produces aluminum wheels in
St. Mary’s, Ohio, and primarily supplies Ford. Enkei America, located in Columbus, Indiana,
mainly supplies Honda with aluminum wheels.
Figure 1 compares the costs of producing aluminum wheels in the United States and in Mexico.
It is based on estimates provided by Lilley Associates for total production costs, as well as the
share accounted for by various costs for a typical high-volume 17-inch wheel sold to the Detroit
3 carmakers.
The largest single cost of producing the wheels is the aluminum. According to American Metal
Market, a 17-inch aluminum wheel contains roughly 20 pounds of aluminum. Lilley Associates
suggests that aluminum accounts for around one-fourth of wheel production costs. According to
DataMonitor, the principal supplier of alumina for the aluminum wheels is Alcoa, with 58.6
percent of the U.S. primary aluminum market in 2008.
112
Canadian Autoparts Toyota only produces cast aluminum wheels; Central Motor Wheels makes both cast and
steel wheels.
224
Table 1: Comparing aluminum wheel production costs between Mexico and
the United States
Materials
Processinga
Casting
Heat treatment
Machining
Painting
SG&Ab
Profit
Other
Total
MEXICO
Percent
30
41
16
18
7
3
1
4
1
11
11
5
5
14
14
8
8
4
9
16
100
2
4
7
44
4
7
13
100
2
4
7
54
$
UNITED STATES
Percent
$
13
24
52
28
20
11
a All processing functions are assumed to be equally labor intensive.
b SG&A stands for selling as well as general and administrative expenses.
Source: Lilley and Associates
Alumina (also known as aluminum oxide) is sourced primarily within the United States for
production of wheels in Mexico as well as in the United States. Alcoa’s long-standing North
American alumina facility is at Massena, New York. A plant at Rockdale, Texas, may also be a
source of primary metal for the Chihuahua plants. Thus, a “Mexican” wheel is likely to include
some U.S. content.
In our example, the $10 (or 22 percent) cost advantage to Mexico originates with the processing
of the alumina, especially casting, machining, and painting as the processing operations are the
most labor-intensive elements of the wheel production process. Yet Mexico no longer represents
the largest source of aluminum wheel imports to the U.S.113
China accounts for an increasing share of global wheel production (as well as U.S.
consumption), according to the trade data. According to Research in China (2008), total
production of aluminum wheels in China has increased from 3.5 million in 2001 to 35 million in
2008. The 35 million figure includes an estimated 15–20 million for motorcycles, 1.5 million for
the aftermarket, and 1.5 million in inventory, leaving 12–17 million for light and heavy motor
vehicles. Per Richard Lilley, about half of the 12 million aluminum wheels imported by the
United States from China in 2008 represent OEM wheels. In the same year, Mexico exported just
under 4 million aluminum wheels to the U.S. Presumably most of these are OEM wheels.
We do not have authoritative information on production costs in China. While manufacturing
wage rates in China are substantially lower than in Mexico, we don’t have a basis for comparing
productivity which would allow us to estimate processing costs in China. We assume them to be
113
See Watkins (2006) on the challenges China represents to manufacturing in Mexico.
225
lower in China than in Mexico because of China’s lower labor costs. China’s aluminum wheels
producers obtain materials from Chinese alumina sources.114
To reach the U.S. market, Chinese wheels incur additional shipping costs compared to
production in Mexico. A standard 40-foot shipping container can hold around 1,000 17-inch
wheels and costs around $1,100 to ship from China to the United States (Huber et al. 2009).
Therefore the shipping cost from China amounts to just over $1 per wheel. While the additional
shipping expense per wheel is likely much less than the labor cost savings on processing,
production in China also triggers incremental inventory costs to make up for the greater distance
to the customer.
Role of Imports in Assembly of Intermediate Part: Seat Assembly
This section demonstrates the complexity of supply chains in the motor vehicle industry and
what this means for identifying the extent to which the import of intermediate goods is hidden.
The seat provides a good example of the challenges in distinguishing between domestic and
foreign sources for motor vehicle parts.
Seats are produced at two types of plants:
1) Plants that specialize in individual parts such as frames, cloth, and foam.
2) Seat assembly plants that assemble seat parts into finished seats ready for installation in
vehicles.
Seat assembly plants are located extremely close to the carmakers’ final assembly plants,
normally within one hour. Seats are delivered to carmakers’ final assembly plants on a just-intime and in-sequence basis, minutes before they are actually installed in the vehicles. Suppliers
assemble seats in response to specific orders from the carmakers; the seats are placed in delivery
trucks in such a manner as to facilitate unloading in the sequence needed on the final assembly
line.
A carmaker’s final assembly plant typically obtains all of its seats from a single seat assembly
plant, and a seat supplier in turn typically dedicates a single facility to producing seats for only
one final assembly plant. Because carmakers have clustered their final assembly plants in auto
alley, so have seat assemblers. Therefore, one might conclude that a seat is a good example of a
domestically produced intermediate part.
However, a closer look at the supply chain reveals that many of the parts that go into making a
seat are actually produced in other countries. A seat consists of several distinct pieces, including
foam padding, leather or fabric, a metal frame, and controls for seat position and temperature.
Most of the parts that go into seats involve straightforward labor-intensive tasks such as cutting
and sewing. Plants producing seat parts do not have to be near seat assembly plants, and instead
can locate in low-wage countries.
114
GM struck a deal to obtain Chinese alumina a few years ago at a favorable price, but assigning a market price to
Chinese alumina is not easy.
226
Trade data illustrate the challenge in identifying the extent of intermediate goods imports (Figure
6). The import of assembled seats is minimal, and is accounted for primarily by a Lear Corp. seat
assembly plant in Windsor, Ontario, that delivers finished seats to a GM final assembly plant
only 10 miles away, but on the other side of the Canada-U.S. border, in Hamtramck, Michigan.
Other than Hamtramck, carmakers’ final assembly plants in the United States receive finished
seats from seat assembly plants also in the United States; therefore, the finished seat is
considered a U.S.-made component.115
Meanwhile, between 1989 and 2007, import of seat parts increased from $621 million to nearly
$5 billion. As U.S. motor vehicle production started to decline sharply toward the end of 2008,
imports of seat parts fell to $4.1 billion in 2008. Imports of seat parts are destined for seat
assembly plants. According to trade data, Mexico accounted for around $2.8 billion of the $4.1
billion imports of seat parts in 2008, and Canada nearly $667 million (Figure 7).116 Thus, we can
conclude that Mexico is a large producer of intermediate goods for U.S. seat assembly.
The example of seat assembly demonstrates that an intermediate good itself consists of
intermediate goods, many of which can be imported. Such supply chain relations may not be
currently reflected in the way import price indices are calculated.
SUMMARY
This paper tries to shed some light on the measurement issues related to growing globalization
by illustrating the complexities of the supply chain of the automobile industry. The production of
automobiles is a large and complex undertaking that involves nearly every manufacturing
industry. Assembly of motor vehicles and production of parts represented 6.5 percent of all U.S.
manufacturing jobs in 2008.
Imports of motor vehicle parts to the United States have been rising as supply chains
increasingly extend across borders. Rising imports of vehicle parts both substitute for U.S.-based
parts production and complement U.S.-based vehicle production of foreign producers. During the
last 15 years the mix of source countries has changed considerably. In particular the share of
imports of intermediate parts from low-wage countries has increased. This paper provides some
background on these trends. The shift of production of aluminum wheels to Mexico as well as
the sourcing of seat parts represent two specific examples discussed in more detail.
115
Two companies dominate production of finished seats in the United States—Lear Corp. and Johnson Controls,
Inc. (JCI). Each has roughly 40 percent of the North American market. Faurecia, Magna, and Trim Masters hold
much of the remaining share.
116
Mexico’s share of seat part imports has averaged 70 percent since 2000.
227
Figure 1: U.S. Light vehicle production (million units) and motor vehicle parts imports
($ bn)
Source: USITC Dataweb, Federal Reserve Board via Haver Analytics
Note: trade data are of annual, light vehicle production data are of quarterly frequency.
Figure 2: U.S. motor vehicle parts imports by major source countries
Source: USITC dataweb, author’s calculations
228
Figure 3: U.S. motor vehicle parts imports by high-wage and low-wage countries
Source: USITC dataweb, authors’ calculations
Figure 4: Largest high-wage and low-wage source countries for US auto parts imports
229
Source: USITC Dataweb and authors’ calculations
Note: Shares are out of all imports
Figure 5: Motor vehicle part imports from Mexico by major subsystems
230
Figure 6: U.S. Imports of automotive seats and seat parts
Source: USITC Dataweb, and authors’ calculations
Figure 7: U.S. Imports of seatparts by largest source country
231
References
Carillo, Jorge, and Oscar Contreras. 2007. “The Historical Evolution of American Auto Firms in
Mexico. Two Models of Local Development: From Local Supplier Networks to Firm
Upgrading.” Paper presented at GERPISA International Colloquium, held in Paris, June
20–22.
Data Monitor. 2008.Aluminum in China. Reference Code 0099-2004. Data Monitor, New York.
Gilmer, Bill, and Jesus Canas. 2008. “The Geography of Cyclical and Structural Change in
Mexico’s Maquiladora Sector: 1990–2006.” Federal Reserve Bank of Dallas, photocopy.
Huber, Thomas, Hans-Werner Kaas, Asutosh Padhi, and Jose Maria Rancano. “ ‘Offshoring’
Onshore: Exploring Mexico’s Manufacturing Advantages.” McKinsey Automotive &
Assembly Extranet.
http://forummexique2009.files.wordpress.com/2009/08/exploringmexicosmanufacturingad
vantages.pdf (accessed January 14, 2009).
Herzenberg, Stephen A. 1991. “The North American Auto Industry at the Onset of Continental
Free Trade Negotiations.” U.S. Department of Labor Economic Discussion paper No. 38.
Washington, DC: U.S. Department of Labor.
Ito, Keiko, Masaru Umemoto. 2004. “Intra-Industry Trade in the ASEAN Region: The Case of
the Automotive Industry.” ASEAN-Auto Project No.04-8, Working Paper Series Vol.
2004-23. Kitakyushu, Japan: International Centre or the Study of East Asian Development.
Klier, Thomas, and James Rubenstein. 2008. Who Really Made Your Car? Restructuring and
Geographic Change in the Auto Industry. Kalamazoo, MI: W.E. Upjohn Institute for
Employment Research.
Montout, Sylvie, Jean Louis Mucchielli, and Soledad Zignago. 2007. “Horizontal and Vertical
Intra-Industry Trade of NAFTA and MERCOSUR: The Case of the Automobile
Industry.” Photocopy, University of Paris.
Research in China. 2008. Global and China Automotive Aluminum Wheel Industry Report,
2008—2009. Report 1082. Beijing: Research in China.
Sturgeon, Timothy, Johannes Van Biesebroeck, and Gary Gereffi. 2007. “Prospects for Canada
in the NAFTA Automotive Industry: A Global Value Chain Analysis.” Paper prepared
for Industry Canada.
U.S. Congress, Office of Technology Assessment. 1992. “U.S.-Mexico Trade: Pulling Together
or Pulling Apart?” ITE-545. Washington, DC: U.S. Government Printing Office.
Watkins, Ralph. 2006. “The China Challenge to Manufacturing in Mexico.” Washington, DC:
U.S. International Trade Commission
233
http://www.offshoregroup.com/newsfiles/chinachallenge.pdf (accessed October, 20,
2009).
Weintraub, Sidney, and Chris Sands. 1998. The North American Auto Industry under NAFTA.
Washington, DC: Center for Strategic and International Studies.
234
CHAIR: Michael Horrigan (Bureau of Labor Statistics)
Bias Due to Input Source Substitutions: Can It Be Measured? Erwin Diewert (University of
British Columbia) and Alice Nakamura (University of Alberta)
Are there Unmeasured Declines in Prices of Imported Final Consumption Goods?, Marshall
Reinsdorf and Robert Yuskavage (Bureau of Economic Analysis)
DISCUSSANT: Barry Bosworth (Brookings Institution)
235
Bias Due to Input Source Substitutions: Can It Be
Measured?
May 2, 2010
W. Erwin Diewert and Alice O. Nakamura117
Discussion Paper 10-07
Department of Economics
University of British Columbia
Vancouver, Canada, V6T 1Z1
Email addresses: [email protected] ; [email protected]
Abstract
Once a business opts to purchase rather than produce an input, it can also change the source from
which the product is procured. Producer price index programs face problems in dealing with
price changes associated with sourcing changes. We present measures for price index bias due to
sourcing substitutions. We begin with highly simplified cases to convey the rationale for our
approach, and then show how the measures could be generalized. We also explain related
aspects of the industry accounts. This material makes it clear that the growth of outsourcing and
the related increases in domestic and foreign sourcing substitutions pose important challenges for
statistics agencies.
117
Department of Economics, University of British Columbia and the University of Alberta School of Business.
This is a revised version of a paper presented at the Conference on “Measurement Issues Arising from the Growth of
Globalization,” sponsored by the W.E. Upjohn Institute and the National Academy of Public Administration and
held in Washington, DC, November 6–7, 2009. The authors thank William Alterman, Robert Feenstra, Mike
Horrigan, Susan Houseman, Emi Nakamura, and Jón Steinsson for helpful comments on various drafts, and the
Social Sciences and Humanities Research Council of Canada (SSHRC) for partial funding. All errors and opinions
are the sole responsibility of the authors.
237
Journal of Economic Literature Classification Numbers
C43, C67, C82, D24, D57, E31, F1
Keywords
Offshoring, outsourcing, consumer price indexes, producer price indexes, intermediate input
price indexes, bias in price indexes, Fisher index, total factor productivity growth, production
accounts, System of National Accounts, exports and imports in the input-output accounts.
238
Producers utilize and make many thousands of products in a year. Index numbers help reduce
and summarize this abundance of microeconomic information. Hence, index numbers intrude
themselves on virtually every aspect of empirical research about firms and the business sector.
The price index programs of the U.S. Bureau of Labor Statistics (BLS) were put in place back
when firms produced on an in-house basis more of the intermediate goods and services required
as inputs. When a firm switches from producing to procuring an input, this is called outsourcing.
Of course, once a firm has found one external source for an intermediate input, it could switch to
another source. Thus, sourcing substitutions should rise in the wake of increased outsourcing.
Unfortunately though, the main producer side measures of inflation, which feed into productivity
and other national economic performance measures, miss much of the price change associated
with sourcing substitutions, and this can cause sourcing substitution bias in the producer price
indexes. This has become an important problem because finding cheaper input sources is a
prevalent modern day business strategy for lowering production costs.
Intermediate input price competition is also believed to play a key role in the survival and growth
of new firms. Foster, Haltiwanger, and Syverson (2008) make this point in an important
empirical study. For firms that produce physically homogenous products, they are able to
evaluate and compare a measure of physical productivity with the more conventional
productivity measure computed using revenue, cost, and price index information. Foster,
Haltiwanger, and Syverson find a strong inverse correlation between physical productivity and
prices that is consistent with newer entrants having lower marginal costs, and with this allowing
those businesses to charge lower prices.118 More generally, they and others document examples
of physically identical producer inputs being available from different sources for different prices
at the same point in time.119
The derivations of our bias formulas in the main body of the paper are easy to understand
because they are for a highly simplified economy. In an appendix, we show that these formulas
can be extended to a more realistic case. Fortunately, the same main parameters emerge as the
determinants of the bias size, though the derivations are more involved.120
We also explain how price index bias problems undermine the validity of the industry accounts
produced by the U.S. Bureau of Economic Analysis (BEA): accounts that support national
productivity measurement and economic policy analysis. We explain how these bias problems
are obscured by and impede efforts to understand the workings of the economy.
118
In another recent AER paper, Bergin, Feenstra, and Hanson (forthcoming) develop a model in which the timing
of when firms begin to offshore a variable-cost activity to foreign firms is endogenously determined in response to
economic conditions, including unit cost price differences for input factors.
119
For example, Byrne, Kovak, and Michaels (2009) report that shifts in the location of production to lower-cost
countries can deliver cost declines of up to 0.8 percent per year. Also, Klier and Rubenstein (2009) report that “The
mass-produced aluminum wheel is a commodity that is sourced by carmakers on the basis of price.” Of course,
firms sometimes purposely buy inputs from sources charging premium prices because of perceived benefits offered
by the higher priced suppliers. In a later section in the paper, “Recommended Data Gap Fixes,” a potential
procedural solution is proposed for dealing with quality differences that might be associated with price changes that
occur together with sourcing substitutions.
120
Houseman et al. (2009) demonstrate the value of this formulation for empirical studies.
239
SETTING THE STAGE
We begin by considering an economy with just enough in it so that we can show how input
source substitution bias arises and can be measured. We consider an economy with four firms,
one truly homogeneous product, and no taxes or transport costs. To explore the importance of
certain price index compilation protocols, we take a case where firms 1 and 2 are suppliers and
firms 3 and 4 are purchasers of the homogeneous product, and where firms 1 and 3 are large and
firms 2 and 4 are initially small, but grow substantially from period 0 to 1. In other words, we
allow for the situation that Foster, Haltiwanger, and Syverson (2008) document where firms
enter that have found ways of producing at lower cost and then gain market share by selling their
products for less than the incumbent firms.
The activities of the four firms can be summarized as follows:
•
Firm 1 is a higher-cost producer that sells exclusively to firm 3 in periods 0 and 1.
•
Firm 2 is a lower-cost producer that is small and only sells to small firm 4 in t = 0 .
However, in t = 1 , firm 2 moves out of the small firm category by selling to firm 4,
which has grown, and by also winning a contract to supply part of the intermediate
product needs of large firm 3.
•
Firm 3 only purchases the output of the higher-cost supplier, firm 1, in period 0, but shifts
some of its purchases to firm 2 in period 1: a sourcing substitution.
•
Firm 4, which is small in t = 0 , only uses the output of the lower cost firm 2 in t = 0,1 .
We let p tjk and q tjk denote the price and quantity for sales of the one product from firm j to firm
k in period t = 0,1 .121 The firm value flows in Table 1 have a positive sign for outputs and a
negative sign for inputs. Firm 1 is always the higher-cost supplier. Thus we always have
(1)
0
p13
> p 024 > 0 ; p113 > p124 > 0 ; and p113 > p123 > 0 .
Table 1. Value Flows between the Four Firms
Output flows
Input flows
Firm 1
Firm 2
Firm 3
Period 0 Value Flows
0 0
p 13
q 13
p 024 q 024
0 0
− p13
q13
Firm 4
− p 024 q 024
Period 1 Value Flows
p113 q113
121
p123q123 + p124 q124
− p113 q113 − p123 q123
− p124 q124
Normally we would expect p123 to be close to p124 , but we allow for possible price discrimination by firm 2.
240
It is important to understand some specifics of the price collection and index compilation
practices of the BLS. The Producer Price Index (PPI) measures inflation in the prices of the
outputs of domestic producers: products sold to other domestic businesses as intermediate inputs
as well as final demand outputs. The Import Price Index (MPI) measures inflation for the
products imported by U.S. businesses and residents. The Export Price Index (XPI) measures
inflation for U.S. export products.122 When a producer changes from buying an intermediate
input from a domestic source to buying that same input from a foreign producer, the product
price will drop out of the domain of definition for the PPI and will become part of the domain of
definition for the MPI, and any price change associated with that sourcing substitution will not
be included in either the PPI or the MPI.
For the PPI and XPI, price collection is on the sellers’ side, whereas for the MPI, price collection
is on the buyers’ side.
As the first step in the procedures for collecting prices from producers, the BLS first selects
establishments. Products are then chosen and prices are collected at the chosen establishments.
The BLS collects price data from establishments roughly in proportion to their transaction
volumes. Prices are collected from some small establishments for both the PPI and the
International Prices Program (responsible for the MPI and the XPI). However, the proportion of
establishments selected for price collection falls steeply with establishment size below a
threshold value for each program.
BLS protocols specify that price collection from businesses should focus on the price forming
units. Firms often centrally determine the selling prices for all of their establishments. Some
price collection is now carried out in a “head office” format, but most still involves interactions
with the establishments where the productive activities are carried out. Of course, it is only for
single establishment firms that the establishment and the firm are the same. Nevertheless, for
convenience, hereafter we will usually refer to the production units as firms.
Abstracting from some of the finer points, the current BLS price index compilation practices for
the PPI and MPI have the consequence that the prices used for index compilation are only for
ongoing supply contracts (see Nakamura and Steinsson [2008, 2009]). The objective of these
BLS practices is to ensure that only identical products, of identical quality, are priced period to
period. This is how the BLS has been implementing the matched model methodology for
comparing price change over time. However, one important reason why firms make sourcing
substitutions is to benefit from lower prices. When firms change the input suppliers they are
using, this inevitably results in the initiation of new supply contracts that, we would think,
usually involve lower prices. Hence, a statistical agency practice of using only the prices from
ongoing supply contracts in index compilation will miss a substantial and systematic source of
price change.
Summarizing the key points from the above discussion, BLS procedures miss price changes that
accompany producer sourcing switches for three main reasons:
122
See Chapter 14 on the PPI Chapter 15 on the MPI and the XPI in the BLS Handbook of Methods, available at
http://www.bls.gov/opub/hom/homtoc.htm . See also Diewert (2007a,b).
241
When a producer switches from a domestic to a foreign supplier (or vice versa), any price change
for the input falls into a price collection gap between the PPI and MPI programs.
In an effort to control for quality differences in products and possible bundled services and
amenities, the BLS typically only uses prices from ongoing contracts for index number
compilation purposes. This practice was instituted to avoid treating quality-related price
differences as true price change.
The lower-cost suppliers that producers switch to are often new, small suppliers trying to gain
market share by undercutting the prices of established firms. Existing BLS price collection
practices mean there is relatively little price collection from small firms, which will tend to result
in missing the lower-price transactions for purchases from newer firms.
We want to focus attention on the above problems that are believed to be the biggest causes of
sourcing substitution bias in the PPI and MPI. Thus, for now, we do not bother with specifying
the relevant BLS program or with whether the price collection is from sellers or purchasers for
the relevant BLS price index program: issues we return to briefly later in the section titled “Price
Collection from Sellers versus Buyers.”
Table 2 shows unit values, given the value flows in Table 1. The unit value for a homogeneous
product is the total cost of all purchases, or the total revenue from all sales, of the product
divided by the number of items transacted.123 Table 3 shows inflation measures for specificed
cases. These inflation measures are ratios of the period 1 to the period 0 unit values for the
different cases in Table 2.
For panel 1 in Tables 2 and 3, only the prices from the sales or purchases of large firms are
included. In contrast, for panel 2, transactions for firms of all sizes are included. Now there are
no blank cells. However, firm 2 sales to firm 3 and firm 3 purchases from firm 2 are still
ignored. In the rest of this paper, we focus on this panel 2 case. (Actual BLS practice falls
between the panel 1 and panel 2 cases in the sense that the BLS does collect prices from some
small firms.)
For panels 3 and 4, all transactions for all firms are included. In panel 3, firm-specific
expressions are shown whereas panel 4 gives the corresponding economy wide aggregate
expressions. We refer subsequently to the expressions in these panels 3 and 4 as true targets.
The “true” or “target” firm indexes in panel 3 of Table 3 can be compared with the incorrectly
measured price indexes given in panels 1 and 2 of Table 3. The entries in Tables 2 and 3 are
used in the following section for showing how the sourcing substitution bias could be measured
123
Since the product being traded between firms is assumed to be homogeneous in all of the following scenarios, the
methodological advice given in the Producer Price Index Manual applies, and unit value prices are the appropriate
prices to insert into index number formulas in deriving the true target price index measures. See IMF et al. (2004,
pp. 509–510), Reinsdorf (1993), and Diewert (1995). The idea that a unit value for homogeneous items is the
appropriate price to use in a bilateral index number formula can be traced back to Walsh (1901, p. 96) (1921, p. 88),
and Davies (1924, 1932). Other index formulas are used in the appendix.
242
when all four firms in the hypothetical economy are domestic. Then in the next two sections, we
extend the analysis to cover cases where some of the firms are foreign.124
124
The “optimal” procedures may simply be too expensive for the agency to implement. But it is good to have our
analysis so that rough estimates of bias could be made.
243
Table 2. Unit Values for Domestic Firms under Alternative Measurement Conditions
Output flows
Period
Input flows
Firm 1
Firm 2
Firm 3
Firm 4
Large in t = 1
Small in t = 1
Large in t = 1
Small in t = 1
1. Firm unit values; based on prices for continiuing contracts at large firms only.
0
1
0 0
p13
q13
no price
0
q13
collection
p113q113
no price
q113
collection
p0 q 0
− 13 13
0
q13
p1 q1
− 13 13
q113
no price
collection
no price
collection
2. Firm unit values; based on prices for continiuing contracts at all firms.
0
1
0 0
p13
q13
p024q 024
0
q13
q 024
p113q113
p124q124
q113
q124
p1 q1
− 13 13
q113
p0 q 0
− 24 24
q 024
p1 q1
− 13 13
q113
p0 q 0
− 24 24
q 024
3. Firm unit values; based on all prices at all firms.
0
1
0 0
p13
q13
p024q 024
0
q13
q 024
p113q113
p123q123 + p124q124
q123 + q124
q113
p0 q 0
− 13 13
0
q13
p0 q 0
− 24 24
q 024
− p113q113 − p123q123
q113 + q123
p1 q1
− 24 24
q124
4. Economy wide unit values; based on all prices at all firms.
0
1
0
q13
+ q 024
⎡ p0 q 0 + p0 q 0 ⎤
24 24 ⎥
− ⎢ 13 13
0
⎥⎦
⎢⎣ q13 + q 024
p113q113 + p123q123 + p124q124
q113 + q123 + q124
⎡ p1 q1 + p1 q1 + p1 q1 ⎤
23 23
24 24 ⎥
− ⎢ 13 13
1
1
1
⎥⎦
⎢⎣
q13 + q 23 + q 24
0 0
p13
q13 + p 024q 024
Note: The unit value expressions shown above are only observable for domestic firms.
244
Table 3. Price Indexes for Domestic Firms under Alternative Measurement Conditions
Output flows
Input flows
Firm 1
Firm 2
Firm 3
Firm 4
Large in t = 0
Small in t = 0
Large in t = 0
Small in t = 0
1. Firm unit values; based on prices for continiuing contracts at large firms only.
p113
no price
p113
no price
0
p13
collection
0
p13
collection
2. Firm unit values; based on prices for continiuing contracts at all firms.
p113
p124
p113
p124
0
p13
p024
0
p13
p024
3. Firm unit values; based on all prices at all firms.
p113
p123q123 + p124q124
p113q113 + p123q123
p124
0
p13
p 024q 024
0 0
p13
q13
p024
4. Economy wide unit values; based on all prices at all firms.
p113q113 + p123q123 + p124q124
0 0
p13
q13 + p024q 024
p113q113 + p123q123 + p124q124
0 0
p13
q13 + p024q 024
Note: The inflation measures shown are only observable for domestic firms.
DOMESTIC SOURCING SUBSTITUTION BIAS
We first take up sourcing substitutions among domestic sources: the most prevalent case. For the
Table 1 value flows, we consider the case for which results are given in panel 2 of Tables 2 and
3.125 We consider the situation in which the only reason some changes in price for the
homogeneous product are missed is because only prices for continuing contracts are used in
index compilation. For reasons expalined later, we focus on the measures for the purchasing
firms 3 and 4 rather than the corresponding measures for the selling firms 1 and 2.
The firm 4 price index that would be computed, denoted here by PI(4) and shown in panel 2 of
Table 3, is compared with the true target index for firm 4, PT(4) , shown in panel 3 of Table 3.
125
In Appendix A, we extend the analysis in this section to the case where we are dealing with the simultaneous
outsourcing of N products instead of just a single product. The methods we use to measure sourcing substitution bias
are similar to, but not the same as, the method used by Diewert (1998) to measure outlet substitution bias in the CPI.
245
Both are simply the ratio of the firm 4 purchase price for the homogeneous product in period 1,
p124 , to its price in period 0, p024 . Thus, for firm 4, we have126
(2)
PT(4) ≡ p124 / p 024 = PI(4) .
The sourcing substitution bias is defined as the measured index minus the true target index. This
difference is zero for firm 4, so there is no sourcing substitution bias problem for this firm.
The analysis for firm 3 is more complex. There are two transactions in period 1 for this firm at
potentially different prices. Since the product being traded is assumed to be homogeneous, as
already noted, the unit value is the appropriate price to use for price (and also quantity) index
compilation. The true firm 3 unit value, u13 , for period 1 is
(3)
u13 ≡ [ p113q113 + p123q123 ] /[ q113 + q123 ] ,
as also shown in panel 3 in Table 2. We find it convenient to define share parameters for the
proportions of firm 1 and firm 2 output sold to firm 3 in period 1; i.e.,
(4)
S113 ≡ q113 /[ q113 + q123 ] and
(5)
S123 ≡ q123 /[ q113 + q123 ] ,
with S113 + S123 = 1 . The unit value expression in Equation (3) can now be rewritten as
(6)
u13 = p113S113 + p123S123 .
Using Eqs. (3)–(6), the target firm 3 price index, shown in panel 3 of Table 3, is
(7)
0
0 1
0 1
PT(3) ≡ u13 / p13
= (p113 / p13
)S13 + (p123 / p13
)S23 .
However, the formula given in Equation (7) is not what would be evaluated for firm 3 if the
period 1 sales of firm 2 to firm 3 are ignored because this is the first period for a new contract.
As given in panel 2, the measured price change component for firm 3 would be
(8)
126
0
PI(3) ≡ p113 / p13
.
We assume that the corresponding true quantity index is obtained by deflating the value ratio by the true price
0 0
0
index. Thus Q T(1) ≡[ p113q113 / p13
q13 ] / PT(1) = q113 / q13
and Q T(4 ) ≡[ p124 q124 / p 024 q 024 ] / PT(4 ) = q124 / q 024 .
246
The numerator in this incorrect index is the period 1 price from the high-cost supplier, p113 ,
rather than the unit value price, u13 , which is an average price for firm 3 input purchases in
period 1 from both the high- and low-cost suppliers.
Before specifying a formula for the bias for the firm 3 price index, it is helpful, for interpretive
purposes, to introduce a bit more notation. We will let i be the rate of price inflation for
deliveries from the high-cost supplier, firm 1, to firm 3.
0
(1 + i) ≡ p113 / p13
= PI(3) ,
(9)
using Equation (8).
Also, we let 0 ≤ d < 1 be a discount factor reflecting the proportional price discount for firm 3
purchases from firm 1 versus firm 2. Thus, we have
(10)
0
(1 + i)(1 − d) = p123 / p13
,
where (1 − d) = p123 / p113 ≤ 1 . From Equation (9), Equation (10), and Equation (7), we now have
(11)
0 1
0 1
PT(3) = (p113 / p13
)S13 + (p123 / p13
)S23
= (1 + i)S113 + (1 + i)(1 − d)S123
= PI(3) − (1 + i)dS123
also using Equation (8) and the property
S113 + S123 = 1 .
Therefore, for the case of a purchasing substituion from one domestic producer to another, the
sourcing substitution bias for firm 3 is defined as the index that is computed, given in Equation
(8), less the true target index given in Equation (7):
(12)
B(3) ≡ PI(3) − PT(3) = (1 + i)dS123 > 0 .
From Equation (12) we see that the sourcing bias is positive since d > 0 , and the product of
three factors:
0
;
1) The rate of price inflation for the high-cost supplier; i.e., (1 + i) ≡ p113 / p13
2) The proportional cost advantage of the low-cost supplier over the high-cost supplier; i.e.,
0
0
) /( p113 / p13
)] > 1 ; and
d = 1 − [( p123 / p13
247
3) The share of deliveries to sector 3 in period 1 that are from the new low-cost supplier; i.e.,
S123 ≡ q123 /[ q113 + q123 ] .
If rough guesses can be made for the cost advantage of the low-cost supplier and the input shares
displaced, then a rough approximation of the bias in the intermediate input price index could be
made using Equation (12).127
We conclude this section with some observations regarding two extensions of the above results
for domestic sourcing substitutions: 1) an extension to the whole economy level, and 2) an
extension to the case where the higher cost supplier, firm 1, shuts down in period 1 because of
the competition from firm 2, so p113 and q113 are not available.
Regarding the whole economy extension, since the firms 1 and 2 are selling a homogeneous
commodity, the true period 1 unit value for the combined economy-wide inputs of firms 3 and 4
is u1 ≡ −[ p113q113 + p123q123 + p124q124 ] /[ q113 + q123 + q124 ] , and the corresponding period 0 whole
0 0
0
economy unit value is u 0 ≡ −[ p13
q13 + p 024q 024 ] /[ q13
+ q 024 ] , as in panel 4, Table 2. Thus the
target price index is PT(3+ 4) ≡ u1 / u 0 , which equals the expression in panel 4 of Table 3.
If the statistical agency fails to include the price information for the new period 1 contract that
firm 2 has to supply firm 3, then the computed economy-wide index would be
0 0
[ p113q113 + p124q124 ] /[ p13
q13 + p024q 024 ] , which will yield a higher value than the target index.
Thus the computed index for the combined firms will have an upward bias.
Moving on to the second extension, suppose firm 1 shuts down in period 1 because of the
competition from firm 2, so that p113 and q113 are not available. Now it is not possible to define
an observable true output price index for firm 1 in period 1. However, the true input price index
for firm 3 can still be defined by Equation (7) with S113 = 0 and S123 = 1 , so PT(3) could be
0
specified to be the (unmatched) price ratio p123 / p13
, and the rest of the algebra above goes
through.
0
, could be evaluated, note that
However, although the true input price index for firm 3, p123 / p13
the period 0 and period 1 price observations come from different suppliers. Given currently
accepted practices, a statistical agency would be more likely to measure the price change by
p124 / p024 . With low inflation for the prices charged by any one firm, this price ratio would be
0
close to one and hence generally larger than the mixed price ratio, p123 / p13
. Thus, in general,
127
In Appendix A, we show that the bias formula becomes more complex when we generalize the above one
commodity case to the case of many commodities, but Equation (12) is valuable as a rough approximation to the
bias.
248
the incorrect index will again have an upward bias, though this bias is not given by Equation
(12).128
OFFSHORE SOURCING SUBSTITUTION BIAS
The extension of the analysis in the previous section to cover the case of a domestic firm
switching among foreign suppliers is straightforward. We simply reinterpret firms 1 and 2 as
foreign suppliers. Now none of the column 1 and 2 expressions can be evaluated in Table 2, so
we focus on the prices that can be observed for firms 3 and 4. Inflation measurement for imports
must be carried out using prices collected from purchasers. (Comparability with the results here
is one reason we chose to focus on price indexes for firms 3 and 4 in the previous section too.)
All of the algebra in the previous section can be applied. There are no sourcing substitution bias
problems for firm 4. However, the statistical agency will be likely to compute the incorrect price
index for firm 3 given by Equation (8) instead of the true target index given by Equation (7), for
the same reasons explained above. Thus, the bias Equation (12) is again operative.
Now imports are in the picture, for each firm in the domestic economy, input purchases should
be distinguished by their point of origin so it can be determined whether a purchase should be
classified as a domestic or an imported input. We further discuss this issue in the next to last
section which is on the BEA Industry Accounts.
DOMESTIC TO FOREIGN SOURCING SUBSTITUTIONS
Finally, we take up the case where firm 1 is a domestic supplier and firm 2 is a lower-cost
foreign supplier of the same homogeneous product. Now it is just the column 2 expressions in
Tables 2 and 3 that cannot be evaluated. However, since firms 3 and 4 are domestic, the column
3 and 4 expressions can still be evaluated. The sourcing substitution for firm 3 is now the switch
from exclusively using a domestic supplier (firm 1) in period 0 to importing some of the firm’s
requirements for the homogeneous input from the foreign firm 2 in period 1.
All of the algebra in section 3 for firm 3 remains valid. Thus the bias analysis in the third section
carries over to the present context in a straightforward manner.
PRICE COLLECTION FROM SELLERS VERSUS BUYERS
With the producer price indexes currently produced by the BLS, the price collection is on the
purchasers’ side for the MPI since it is only the purchasers that are domestic, but it is on the
sellers’ side for the PPI. We ignored this institutional information in the previous sections and
proceeded as though all price collection were on the buyers’ side. We did this partly for
expositional convenience, but also because doing this allowed us to demonstrate that the
128
0
The new bias formula will be B≡ PI(3) − PT(3) = (p124 / p 024 )− (p123 / p13
).
249
sourcing substitution bias problem cannot be corrected solely by creating a comprehensive input
price index with price collection on the purchasers’ side.129
As of now, the United States does not have a comprehensive input price index. Rather, the
components of the PPI program that are for intermediate products sold to other businesses and
the price index components of the MPI for purchases by businesses are combined to form a
pseudo input price index. The BLS hopes to begin producing a true Input Price Index (IPI) in the
future, as explained in Alterman (2009), but has not yet been given the funding to do so. We
view this as an important step forward. However, to deal with the sourcing substitution bias
problem, the BLS also needs to utilize the initial observation when a new procurement contract is
started, relating this price to the price paid under the previous contract, and to collect prices from
all firms rather than mostly from large ones. The purchaser knows, presumably, whether the
product is truly the same despite a price change.130 On the other hand, a seller typically has no
information about the prior purchasing choices of their customers or their reasons for making
sourcing changes. So, new sources of supply should be flagged and the statistical agency must
decide, case by case, on whether the product from a new source of supply is closely comparable
to the product from the old source of supply. This is the information that would make the BLS
feel comfortable about including prices for first time purchases under new procurement contracts
in producer price index calculations. Making these changes will not be easy, but our bias
Equation (12) makes it clear the changes are important!
THE BEA INDUSTRY ACCOUNTS
One important use of producer price indexes is for use by the BEA for producing industry
accounts that are corrected for pure price change over time. The BEA industry accounts, which
include the benchmark input-output (I-O) accounts and annual GDP-by-industry accounts, are
heavily used for economic policy analyses.131 Benchmark I-O accounts are produced every five
years using data from the quinquennial economic census of businesses. These include make, use,
and requirements tables. A make table shows the value of each product produced by each
industry in a given year.132 A use table shows the uses of products by intermediate and final
users.133
Starting with the initial benchmark make and use tables, the BEA produces balanced make and
use tables using a procedure that sequentially changes row and column figures until the recorded
129
Almost 50 years ago, the 1961 report of the NBER Price Statistics Review Committee (the Stigler Report) also
recommended that the BLS should rely on purchasers’ prices rather than the sellers’ prices.
130
If the purchaser is acting on behalf of another party, then the question might need to be referred to that other
party.
131
See, for example, Streitwieser (2009) on the ongoing sorts of public policy uses of the BEA industry accounts.
132
Each industry gets a row in a make table, and each product gets a column. The values in a row sum to the current
value of output for the stated industry. Each column total equals the total output for the given product.
133
Each product gets a row, and each row sum is the gross output for the given product. There are columns for the
industry intermediate uses and columns for final uses. The value added for an industry is the sum of the values of
total industry sales (gross output) minus the value of purchases from other industries (aggregate intermediate
industry input). Industry value added summed over industries equals the business sector GDP for the nation.
250
total use of each product equals its recorded total supply, and the recorded final demands equal
the values given in the U.S. national income and product accounts (NIPAs).134
Requirements tables are derived from the balanced make and use tables. The coefficients in the
requirements tables represent interindustry linkages that, in turn, link output and final demand.
For example, the entries in the employment requirements table show the estimated direct and
indirect impacts of a change in final demand on employment in the different industries.
Requirements tables are used for policy impact analyses.
For the purpose of producing constant price real I-O accounts from nominal I-O accounts, price
indexes are needed for deflating the outputs and also the inputs. Unfortunately, as explained
above, we believe the input price indexes for producers are distorted by sourcing substitution
biases. In addition, there are gaps in intermediate product price collection. Imputed rather than
direct pricing is used for numerous products. Moreover, the same economy-wide producer price
deflator is generally used to deflate the value flows between and from each of the industries.
Diewert (2007c) notes that this procedure is correct only if each industry produces the same mix
of micro commodities within each of the broad commodity classes and micro commodity prices
move together across industries: conditions unlikely to be satisfied for a nation!
Moving to the annual industry accounts (AIAs), these consist of the annual GDP-by-industry
accounts and the annual I-O accounts. The annual GDP-by-industry estimates are based on
annual survey and administrative data from several sources. In contrast, the annual I-O accounts
are produced by updating the benchmark I-O accounts utilizing the assumption that the real
(constant-price) use of intermediate inputs relative to each industry’s real gross output does not
change year-to-year; see Strassner, Yuskavage, and Lee (2009). The estimates of an industry’s
real intermediate inputs are first updated based on projected changes in the industry’s real gross
output from the GDP-by-industry accounts and using the observed ratios of intermediate input
use to gross output. The AIAs are then “integrated” internally and with the NIPAs. The
integration achieved is referred to as partial integration.
The balancing process can be viewed, Parker (2004) suggests, as using the relative strengths of
the available data to produce the best possible data for users. However, an initial imbalance, say,
in the initial use table in the form of too little, or too much, supply of a product after the
estimated intermediate input and final uses have been accounted for represents either an
inconsistency in the data or in the framework for the accounts. The balancing process renders
less visible the initial imbalances; see Meade (2006).
Filling the intermediate products value and price data gaps would help the BEA to move toward
full rather than just partial integration of the industry accounts. Lawson, Moyer, and Okubo
(2006) note that the methods developed by the BEA to achieve partial integration were never
seen as an adequate substitute for the needed improvements to the source data. They explain
that, with full integration, the measures of value added by industry would be independent of the
NIPAs and could provide a useful “feedback” loop that would improve the estimates of the
product composition of GDP.
134
The techniques used for this balancing are explained in Appendix B of Lawson, Moyer, and Okubo (2006). See
also Horowitz and Planting (2006) and Yuskavage, Strassner, and Medeiros (2008).
251
Also, policymakers would like to know what factors account for current U.S. offshoring. They
would like to understand and be able to foresee and perhaps influence the major effects of
offshoring on U.S. workers and the economy.135 The I-O tables serve as the framework for
combining the available data for estimated GDP, and are essential to empirical studies of how
outsourcing and offshoring and inshoring are affecting the U.S. economy. However, to properly
allow for foreign engagement, the commodity classification that is used in the I-O tables must be
expanded. A gross output that is being produced by a particular industry in a particular
commodity category must be further distinguished as being supplied to the domestic market or as
an export while an intermediate input that is being used by a particular industry in a particular
commodity category must be further distinguished as being purchased from a domestic or a
foreign supplier. Making these changes would not be a dramatic methodological leap since the
1993 System of National Accounts (SNA93) guidelines already suggested this treatment of
intermediate inputs as a supplementary table.136 However, this change would be expensive.
If it is deemed important to have information on the exports produced and the imports demanded
by each industrial sector, it will be necessary to construct unit value output prices for both
exports and deliveries to domestic demanders and to construct unit value input prices for both
imports and deliveries to the sector from domestic suppliers. If we take this approach as the
ideal, the dimensionality of the supply and use tables would be expanded beyond what could
conceivably be implemented in terms of needed data collection, given present survey data
collection methods and business concerns about confidentiality. However, the suggested
approach could be partially implemented at least, as suggested by Diewert (2001).137
If our purpose is to measure industry productivity, or to measure industry level product or labor
demand impacts, then the answer is reasonably straightforward. When calculating the constant
dollar I-O matrices, each value cell for outputs and each value cell for inputs needs to be deflated
by a price index that matches up with the value flows in that cell. At present, however, there
simply are no adequate surveys on the interindustry flows of services. Even in manufacturing,
where information on commodity flows is relatively complete thanks to explicit surveys of
manufacturing industries, relatively little information on the flows of purchased services is
collected. More attention also needs to be given to the development of basic prices by industry
and by commodity—i.e., we need accurate information on the exact location of indirect taxes
(and commodity subsidies) by commodity and industry on both outputs and inputs.
135
See Dey, Houseman, and Polivka (2006); Freenstra and Bradford (2009); Jarmin, Krizan and Tang (2009);
Kletzer (2009); and Norwood et al. (2006).
136
See Table 15.5 in Eurostat et al. (1993). For a more detailed discussion of how exports and imports could be
introduced into the production accounts, see Diewert (2007a,b) and IMF et al. (2009).
137
See Diewert (2005, 2007c) for a treatment of these problems in a closed economy context, and Diewert (2007a,b)
for an open economy treatment.
252
RECOMMENDED DATA GAP FIXES
As explained following Table 1, the present BLS price collection procedures miss price changes
of three sorts that accompany producer sourcing switches: 1) price changes associated with
domestic to foreign sourcing substitutions (or vice versa), 2) price changes associated with the
start of new procurement contracts, and 3) price changes involving sales of or purchases from
small firms for one of the two time periods involved for the price change. The resulting price
data gaps lead to sourcing substitution bias problems.
Changes in statistical agency practices could fix these data-based problems, though these are
changes that a statistical agency like the BLS would need additional funding to implement. A
comprehensive Input Price Index (IPI) is needed of the sort the BLS would like to have the funds
to produce. Price collection should be from purchasers rather than sellers whenever possible,
since it is only the purchasers who are in a position to state whether there were quality changes
as well as price changes associated with sourcing changes. Also, it is important for data to be
collected from all firms, or at least from higher proportions of small firms.
We believe that an important step toward achieving the needed expansion of data collection from
firms is for the present survey data collection methods to progressively be replaced with full
electronic price and value data capture for all transactions of all businesses large enough to have
electronic information systems.138 Many firms have taken advantage of the low cost of computing
and now have detailed data on all their financial transactions (e.g., they have the value of each
sale and the quantity sold by commodity).139 This information could enable industry and firm
research like the scanner data studies that have proven to be so useful in the context of the
Consumer Price Index.140 This information would permit the construction of true microeconomic
price and quantity indexes at the firm level and the evaluation of more accurate firm and industry
productivity indexes. The 2008 American Economic Review study of Foster, Haltiwanger, and
Syverson makes it clear that having this sort of information could result in a revolutionary
rewrite of our understanding of how productivity growth takes place in the U.S. economy.
Some additional tentative conclusions also follow from our analysis:
•
Basic index number theory and statistical agency practice has not paid enough attention
to the problems that arise when new firms enter and some grow, and some established
firms shrink and then exit.
•
It seems likely that statistical agency operating procedures have led to upward sourcing
substitution biases in the intermediate product components of the PPI and MPI. Upward
biases in input price indexes lead to downward biases in the corresponding quantity
indexes and have the effect of overstating total factor productivity improvements.
138
Abraham and Spletzer argue convincingly that economies of scale would be realized if data were collected, when
possible, for entire firms from their head offices. The context in which Abraham and Spletzer (2009) present these
options is different, but the options they outline seem relevant for our context too.
139
For more on new ways of collecting price data, see also Gudmundsdottir, Gudnason, and Jonsdottir (2008) and
Grimm, Moulton, and Wasshausen (2002).
140
See, for example, Reinsdorf (1994a,b); Nakamura (2008), Ivancic, Diewert, and K.J. Fox (forthcoming); and
Nakamura, Nakamura, and Nakamura (forthcoming).
253
•
Since the value of international trade as a proportion of GDP has mostly increased over
time, it seems likely that sourcing substitution bias has also increased over time.
•
There are other data gaps in the U.S. economic statistics that compound the problems
resulting from sourcing substitution bias. In particular, currently in the United States,
value information for intermediate products is only collected every five years, and there
is no direct price collection for large numbers of intermediate products of economic
importance.
N. Gregory Mankiw (2006, p. 44) quotes John Maynard Keynes as stating:
“If economists could manage to get themselves thought of as humble, competent people
on a level with dentists, that would be splendid.”
Mankiw explains that Keynes was expressing a hope that macroeconomics would evolve into a
useful sort of engineering and that “avoiding a recession would be as straightforward as filling a
cavity.” He goes on to lament a paucity of those in our profession who are willing to commit
their time and energies to helping to achieve the objectives that Keynes enunciated. However,
dentists work on the teeth of their patients with the aid of lights, special mirrors, patients who
willingly open their mouths, and dental X rays. In our view, for economists to make greater
progress on understanding the economy and for economists and government policymakers to be
able to do better on managing the economy, we need better “equipment” for seeing the economy.
The data improvements recommended in this article would help in this regard.
254
Appendix A: Outsourcing Bias when there are N Commodities Being Outsourced
The analysis in the main text associated with Table 1 deals only with the case of a single
homogeneous product. In this appendix, we define indexes as aggregates over N products instead
of a single product. The overall message remains the same but the details are more complex.
In this more general setup, there are four groups of firms. We will assume that group 3
simultaneously switches their sourcing arrangements for N homogeneous commodities. In
particular, group 3 switches some of its input procurement contracts for N commodities group 3
firms need as inputs from group 1 (the higher-cost supplier) to group 2 (the lower-cost supplier)
for N commodities. Thus, the flows shown in Table 1 are still applicable except that now each
price and quantity shown in the table is interpreted as a vector and the old ordinary price and
quantity products must now be interpreted as inner products of the corresponding price and
quantity vectors. Thus, the old value flow of supplies from group i to group j in period t, p ijt q ijt ,
t
t
t
is replaced by p ijt ⋅ q ijt ≡ ∑ nN=1 p ijn
q ijn
for t = 0,1 where now p ijt = [ p ijt 1 , p ijt 2 ,K , p ijN
] is a price
t
vector and q ijt = [ q ijt 1 , q ijt 2 ,K , q ijN
] is a quantity vector.
Let p t and q t be generic price and quantity vectors pertaining to a group for periods t = 0,1 .
Then the Laspeyres and Paasche price indexes, PL and PP , are defined as follows:
(A1) PL (p 0 , p1 , q 0 , q1 ) ≡ p1 ⋅ q 0 /
The Fisher (1922) ideal price index,
Paasche price indexes:
(A2)
p 0 ⋅ q 0 and PP (p 0 , p1 , q 0 , q1 ) ≡ p1 ⋅ q1 / p 0 ⋅ q1 .
PF , is defined as the geometric mean of the Laspeyres and
PF (p 0 , p1 , q 0 , q1 ) ≡ [ PL (p 0 , p1 , q 0 , q1 )PP (p 0 , p1 , q 0 , q1 )]1 / 2 .
The Fisher price index can be justified as a “best” index from multiple perspectives141 and will be
used for forming the “true” target index in what follows.
Looking at Table 1, it is straightforward to define the true output price index for group 1, PT(1) ,
0 1
0 1
as the Fisher index PF (p13
, p13 , q13
, q13 ) and the true input price index for group 4, PT(4) , as the
usual Fisher index PF (p 024 , p124 , q 024 , q124 ) . There are no sourcing substitution complications here.
However, in period 1, group 2 firms began selling the same product to group 3 as well as group 4
firms, possibly at different prices. Thus, proper measurement of inflation for group 2 and group 3
transactions is more complicated.
For groups 2 and 3, the unit value prices for the N commodities ( n = 1,K , N ) in period 1 are
141
See Diewert (1976, 1992) and the Producer Price Index Manual.
255
(A3)
u12n ≡ [ p123n q123n + p124n q124n ] /[ q123n + q124n ] = p123n S123n + p124n S124n ;
(A4)
u13n ≡ [ p113n q113n + p123n q123n ] /[ q113n + q123n ] = p113n S113n + p123n S123n ;
where the group 2 and 3 (physical) quantity shares for product n in period 1 are given by
(A5)
S123n ≡ q123n /[ q123n + q124n ] ; S124n ≡ q124n /[ q123n + q124n ] ; S123n + S124n = 1 ; and
(A6)
1
1
1
1
1
1
1
1
1
1
S13
n ≡ q13n /[ q13n + q 23n ] ; S23n ≡ q 23n /[ q13n + q 23n ] ; S13n + S23n = 1 .
Let the vector of group 2 unit value prices for period 1 be u21 ≡ [u211,u221,...,u2N1] where the u2n1
are defined by (A3) and let the vector of group 3 unit value prices for period 1 be u31 ≡
[u311,u321,...,u3N1] where the u3n1 are defined by (A4). The period 1 quantity vectors that
correspond to these unit value vectors in period 1 are q231+q241 for group 2 and q131+q231 for
group 3. Thus our true output index for group 2 and our true input index for group 3 are defined
to be the following Fisher ideal price indexes, respectively,
(A7)
(A8)
PT(2) ≡ PF (p 024 , u12 , q 024 , q123 + q124 )
= [ PL (p 024 , u12 , q 024 , q123 + q124 )PP (p 024 , u12 , q 024 , q123 + q124 )]1 / 2 ;
0 1 0 1
PT(3) ≡ PF (p13
, u 3 , q13 , q13 + q123 )
0 1 0 1
0 1 0 1
= [ PL (p13
, u 3 , q13 , q13 + q123 )PP (p13
, u 3 , q13 , q13 + q123 )]1 / 2 .
In principle, a statistical agency could compute the true output price index PT(2) defined by (A7)
and the true input price index, PT(3) defined by (A8). In practice, there are likely to be problems.
Of particular relevance here, group 3 has switched to a new supplier in period 1 for the N
commodities under consideration, so there will be no matching price for these supplies in period
0. Given current BLS practices, it is likely that the statistical agency will use the following
“matched model” incorrect intermediate input price index for group 3:
(A9)
0 1 0 1
0 1 0 1 1/ 2
PI(3) ≡ [ PL (p13
, p3 , q13 , q13 )PP (p13
, p3 , q13 , q13 )] .
The incorrect index PI(3) is the geometric mean of the Laspeyres and Paasche price indexes,
0 1 0 1
0 1 0 1
, p3 , q13 , q13 ). and PP (p13
PL (p13
, p3 , q13 , q13 ) , that use only the price and quantity data for the
incumbent supply source (group 1 firms) for both periods. Since group 1 is a high-cost supplier,
we can expect PI(3) to be higher than the true index, PT(3) . Here we will develop formulae that
will enable us to determine the magnitude of this upward bias.
256
It is cumbersome to develop a bias formula for the difference of PI(3) less PT(3) but it is fairly
easy to develop bias formulas for the differences between the Laspeyres and Paasche
components of these indexes. Thus, we start our analysis by expressing the high-cost supplier
0
price relatives, p113n / p13
n , as the following high-cost supplier product specific inflation rates:
0
(A10) p113n / p13
n ≡ (1 + i n ) ;
n = 1,K , N ;
Next, the lower-cost supplier prices in period 1 relative to the corresponding high-cost supplier
prices in period 0 are expressed as:
0
(A11) p123n / p13
n ≡ (1 − d n )(1 + i n ) ,
n = 1,K , N
where d n is a proportional discount factor for the low- versus the higher-cost supplier for
product n that satisfies
(A12) 0 < d n < 1 ;
n = 1,K , N .
(3)
We start our analysis by looking at the Laspeyres component PTL
of the true input price index
defined by (A8)
(A13)
0
0
(3)
PTL
≡ PL (p13
, u13 , q13
, q113 + q123 )
0
0
0
≡ u13 ⋅ q13
/ p13
⋅ q13
0
0 0
= ∑ nN=1 [ p113n S113n + p123n S123n ]q13
n / p13 q13
using definitions (A4)
0
1
1
0
1
0
0
0 0
= ∑ nN=1 [ (p113n / p13
n )S13n + (p 23n / p13n )S 23n ] p13n q13n / p13 q13
1
1
0
0
0 0
= ∑ nN=1 [ (1 + i n )S13
n + (1 − d n )(1 + i n )S 23n ] p13n q13n / p13 q13
using (A10) and
(A11)
0
= ∑ nN=1 [ (1 + i n )S113n + (1 − d n )(1 + i n )S123n ]s13
n
using (A14) below
0
1
0
N
= ∑ nN=1 [ (1 + i n )s13
using (A6)
n − ∑ n =1 d n (1 + i n )S 23n s13n
where the period 0 expenditure shares on the N commodities in group 3 are defined as follows:
257
0
0
0
0 0
(A14) s13
n ≡ p13n q13n / p13q13 ;
n = 1,K , N .
0
It is straightforward to show that ∑ nN=1 (1 + i n )s13
n is the incorrect Laspeyres component,
0 1
0 1
, p13 , q13
, q13 ) , in the incorrect Fisher price index for group 3 defined by (A9) above; so,
PL (p13
0
0
0
(3)
(A15) PIL
≡ PL (p13
, p113 , q13
, q113 ) = ∑ nN=1 (1 + i n )s13
n .
(3)
Thus if we define the bias BL in the incorrect Laspeyres index as the difference between PIL
(3)
and PTL
, using (A13) and (A15), we have the following expression for this bias:
0
(3)
(3)
(A16) B L ≡ PIL
− PTL
= ∑ nN=1 d n (1 + i n )S123n s13
n > 0,
where the inequality follows from the nonnegativity of the physical shares S123n (with at least
0
one of these shares being positive), the positivity of the base period expenditure shares s13
n and
the positivity of the discount factors d n . The bias formula (A16) has the same general structure
as the bias formula (12) in the main text except that now it is the base period expenditure shares
0
s13
n of the group making the N-product sourcing switch that enter the formula.
We now need to repeat the above analysis for the Paasche component of the true index PT(3)
defined by (A8) and the Paasche component of the incorrect index PI(3) defined by (A9). Define
these Paasche components as follows:
(3)
0 1 0 1
0 1
(A17) PTP
≡ PP (p13
, u 3 , q13 , q13 + q123 ) ≡ u13 (q113 + q123 ) / p13
(q13 + q123 ) ;
(3)
0 1 0 1
(A18) PIP
≡ PP (p13
, u 3 , q13 , q13 )
0 1
≡ p113q113 / p13
q13 .
0
In place of the base period expenditure shares s13
n , for our Paasche analysis, we require two sets
of expenditure weights that use the prices of period 0 but quantities that pertain to period 1.
These hybrid expenditure shares are given by:
0
1
0 1
(A19) s 01
n ≡ p13 n q13 n / p13q13 ;
n = 1,...,N ;
*
0
1
1
0
1
1
(A20) s 01
n ≡ p13 n (q13 n + q 23 n ) / p13 (q13 + q 23 ) ;
n = 1,...,N.
01
The expenditure shares s 01
n use the base period prices for group 3, p n , and the deliveries of the
*
use the base
high cost group to group 3 in period 1, q113 , whereas the expenditure shares s 01
n
258
0
, as the price vector and the sum of all deliveries to group 3 in
period prices for group 3, p13
period 1, q113 + q123 , as the quantity vector.
(3)
of the true input price index defined by (A8).
We must now look at the Paasche component PTP
Using definition (A17), we have
( 3)
0
(A21) PTP
≡ u13 (q113 + q123 ) / p13
⋅ (q113 + q123 )
0
= ∑n =1[p113n S113n + p123nS123n ][q113n + q123n ] / p13
[q113 + q123 ]
using (A4)
0
1
1
0
1
01*
= ∑n =1[(p113n / p13
n )S13 n + ( p 23 n / p13 n )S23 n ]s n
using (A20)
*
= ∑n =1[(1 + i n )S113n + (1 − d n )(1 + i n )S123n ]s 01
n
using (A10) and
*
1
01*
= ∑n =1[(1 + i n )s 01
n − ∑n =1 d n (1 + i n )S23 n s n
using (A6).
N
N
N
(A11)
N
N
(3)
of the incorrect input price index defined by
Moreover, for the Paasche component PIP
(A9), using definition (A18), we have
0
⋅ q113
(A22) PIP( 3) ≡ p113 ⋅ q113 / p13
1
0
0
1
0 1
= ∑n =1 (p13
n / p13 n ) p13 n q13 n / p13q13
N
= ∑n =1 (1 + i n )s 01
n
N
using (A10) and (A19).
(3)
(3)
and PTP
.
Define the bias BP in the incorrect Paasche index as the difference between PIP
Using (A21) and (A22), we have the following expression for this bias:
(3)
(3)
− PTP
(A23) BP ≡ PIP
01*
1
01*
= ∑n =1 (1 + i n )s 01
n − ∑n =1 (1 + i n )s n + ∑n =1 d n (1 + i n )S 23 n s13 n
N
N
N
1
01*
01*
= ∑n =1 (1 + i n )[s 01
n − s n ] + ∑n =1 d n (1 + i n )S23 n s13 n .
N
N
259
Under normal conditions, the first term in the last line of (A23) will be close to zero142 so the
N
01*
second term, ∑n =1 d n (1 + i n )S123n s13
n , will dominate. Since this second term is positive under our
assumptions, the Paasche component of the bias, BP , will usually be positive. This second term
has the same general form as the Laspeyres bias component BL defined above by (A16).
We can approximate the true Fisher index PT(3) defined by (A8) by the arithmetic mean of its
Laspeyres and Paasche components. Also, we can approximate the incorrect Fisher index PI(3)
defined by (A9) by the arithmetic mean of its Laspeyres and Paasche components. Expressions
(A16) and (A23) can be used to form an overall bias estimate for the incorrect Fisher index.
142
If all of the commodity specific inflation rates for the high cost producer are equal (i.e., the in are all equal), then
it can be seen that the first term on the right hand side of (A23) will vanish since the two sets of shares sum to one.
This term will also be zero if the correlation between the vector of commodity specific inflation rates in and the
vector of differences in the shares is zero.
260
References
Abraham, K.G., and J.R. Spletzer. 2009. “Addressing the Demand for Time Series and
Longitudinal Data on Occupational Employment.” Presented at the conference
“Measurement Issues Arising from the Growth of Globalization,” held in Washington,
DC, November 6–7.
Alterman, W. 2008. “Globalization and Price Measurement Challenges and Options.”
Powerpoint presentation at the World Congress on National Accounts and Economic
Performance Measures for Nations, held in Washington, DC, May 13–17.
Alterman, W. 2009. “Producing an Input Price Index.” Presented at the conference
DC, November 6–7.
Bergin, Paul R., Robert C. Feenstra, and Gordon H. Hanson. Forthcoming. “Offshoring and
Volatility: Evidence from Mexico’s Maquiladora Industry.” American Economic Review.
Byrne, D., B.K. Kovak, and R. Michaels. 2009. “Offshoring and Price Measurement in the
Semiconductor Industry.” Presented at the conference “Measurement Issues Arising from
the Growth of Globalization,” held in Washington, D.C., November 6–7
Davies, G.R. 1924. “The Problem of a Standard Index Number Formula.” Journal of the
American Statistical Association 19: 180–188.
———. 1932. “Index Numbers in Mathematical Economics.” Journal of the American
Statistical Association 27: 58–64.
Dey, M., S. Houseman, and A. Polivka. 2006. “Manufacturers’ Outsourcing to Employment
Services.” W.E. Upjohn Institute Staff Working Paper No. 07–132. Kalamazoo, MI:
W.E. Upjohn Institute for Employment Research.
Diewert, W.E. 1976. “Exact and Superlative Indexes.” Journal of Econometrics 4: 115–145.
———. 1992. “Fisher Ideal Output, Input and Productivity Indexes Revisited.” Journal of
Productivity Analysis 3: 211–248.
———. 1995. “Axiomatic and Economic Approaches to Elementary Price Indexes.” Discussion
Paper No. 95-01. Vancouver, Canada: Department of Economics, University of British
Columbia.
———. 1998. “Index Number Issues in the Consumer Price Index.” Journal of Economic
Perspectives 12 (1): 47–58.
———. 2001. “Which (Old) Ideas on Productivity Measurement Are Ready to Use?” In New
Developments in Productivity Analysis C.R. Hulten, E.R. Dean, and M.J. Harper, eds.
Chicago: University of Chicago Press, pp. 85–102. www.nber.org/chapters/c10124
(accessed May 25, 2010).
261
———. 2005. “The Treatment of Indirect Taxes and Margins and the Reconciliation of Industry
with National Productivity Measures.” Discussion Paper No. 05-06. Vancouver, Canada:
Department of Economics, University of British Columbia.
http://www.econ.ubc.ca/discpapers/dp0506.pdf (accessed May 25, 2010).
———. 2007a. “Export Import Price Index Manual: The Economic Approach.” Draft chapter.
http://www.econ.ubc.ca/diewert/chapter17.pdf (accessed May 25, 2010).
———. 2007b. “Export Import Price Index Manual: Price Indexes Using an Artificial Data Set.”
Draft chapter. http://www.econ.ubc.ca/diewert/chapter19.pdf (accessed May 25, 2010).
———. 2007c. “Measuring Productivity in the System of National Accounts.” Discussion Paper
No. 07-06. Vancouver, Canada: Department of Economics, University of British
Columbia, V6T 1Z1. http://www.econ.ubc.ca/diewert/dp0706.pdf (accessed May 25,
2010).
Eurostat, IMF, OECD, UN, and the World Bank. 1993. System of National Accounts 1993. New
York: United Nations.
Feenstra, R.C., and J.B. Bradford. 2009. “Evaluating Estimates of Materials Offshoring from
U.S. Manufacturing.” Presented at the conference “Measurement Issues Arising from the
Growth of Globalization,” held in Washington, DC, November 6–7.
Foster, L., J. Haltiwanger, and C. Syverson. 2008. “Reallocation, Firm Turnover and Efficiency:
Selection on Productivity or Profitability?” American Economic Review 98(1): 394–425
http://www.aeaweb.org/articles.php?doi=10.1257/aer.98.1.394 (accessed May 25, 2010).
Grimm, B.T., B.R. Moulton, and D.B. Wasshausen. 2002. “Information Processing Equipment
and Software in the National Accounts.” Paper presented at the conference “Measuring
Capital in the New Economy,” Sponsored by the NBER/CRIW and the U.S. Federal
Reserve Board.
Gudmundsdottir, H., R. Gudnason, and G.R. Jonsdottir. 2008. “Electronic Data Collection
Methods: The Use of Transactions Data for the PPI.” Paper presented at the World
Congress on National Accounts and Economic Performance Measures for Nations, held
in Washington, DC, May 13–17.
Horowitz, K.J., and M.A. Planting. 2006. Concepts and Methods of the Input-Output Accounts,
U.S. Bureau of Economic Analysis. Washington, DC: BEA.
http://www.bea.gov/papers/pdf/IOmanual_092906.pdf (accessed May 25, 2010).
Houseman, S.N. 2007. “Outsourcing, Offshoring, and Productivity Measurement in U.S.
Manufacturing.” International Labour Review 146(1-2): 61–80.
———. 2008. “Outsourcing and Offshoring: Problems for Price and Productivity
Measurement”, paper presented at the World Congress on National Accounts and
Economic Performance Measures for Nations, Washington, DC, May 12–17.
262
———. 2009. “Measuring Offshore Outsourcing and Offshoring Problems for Economic
Statistics.” Employment Research 16(1): 1.
http://www.upjohn.org/publications/newsletter/SNH_109.pdf (accessed May 25,
2010).
Houseman, S.N., C. Kurz, P. Lengermann, and B. Mandel. 2009. “Offshoring Bias: The Effect of
Import Price Mismeasurement on Manufacturing Productivity.” Presented at the
conference “Measurement Issues Arising from the Growth of Globalization,” held in
Washington, DC, November 6–7.
International Monetary Fund (IMF), International Labour Office, Organisation for Economic Cooperation and Development, Eurostat, United Nations, and The World Bank.
2004.Producer Price Index Manual: Theory and Practice (PPI Manual). Chapters and
whole can be downloaded for free at
http://www.imf.org/external/np/sta/tegppi/index.htm.
———. 2009. Export and Import Price Index Manual: Theory and Practice. Washington: IMF.
http://www.imf.org/external/np/sta/tegeipi/ (accessed May 25, 2010).
Ivancic, L., W.E. Diewert, and K.J. Fox. Forthcoming. “Scanner Data, Time Aggregation and the
Construction of Price Indexes.” Journal of Econometrics.
Jarmin, R., C.J. Krizan, and J. Tang. 2009. “Outsourcing, Offshoring, and Trade: Identifying
Foreign Activity across Census Data Products.” Presented at the conference
DC, November 6–7.
Kletzer, L.G. 2009. “Understanding the Domestic Labor Market Impact of Offshore Services
Outsourcing: Measurement Issues.” Presented at the conference “Measurement Issues
Arising from the Growth of Globalization” held in Washington, DC, November 6–7.
Klier, T.H., and J.M. Rubenstein. 2009. “Imports of Intermediate Parts in the Auto Industry—A
Case Study.” Presented at the conference “Measurement Issues Arising from the Growth
of Globalization,” held in Washington, DC, November 6–7.
Lawson, A.M., B.C. Moyer, and S. Okubo. 2006. “Integrating Industry and National Economic
Accounts: First Steps and Future Improvements.” In A New Architecture for the U.S.
National Accounts, D.W. Jorgenson, J.S. Landefeld, and W.D. Nordhaus, eds. Chicago:
University of Chicago Press, pp. 215–262. http://www.nber.org/books/jorg06-1. Chapter
URL: http://www.nber.org/chapters/c0138 (accessed May 25, 2010).
Mandel, M. 2007. “The Real Cost Of Offshoring.” Business Week, June 18 (cover story).
http://www.businessweek.com/magazine/content/07_25/b4039001.htm (accessed May
25, 2010).
Mankiw, N.G. 2006. “The Macroeconomist as Scientist and Engineer.” Journal of Economic
Perspectives 20(4): 29–46.
263
Meade, D.S. 2006. “Why Real Value Added Is Not My Favorite Concept.”
http://inforumweb.umd.edu/papers/conferences/2006/RealValueAdded.pdf (accessed
May 25, 2010).
Nakamura, A., E. Nakamura, and L. Nakamura. Forthcoming. “Price Dynamics, Retail Chains
and Inflation Measurement.” Journal of Econometrics.
Nakamura, E. 2008. “Pass-Through in Retail and Wholesale.” American Economic Review
98(2): 430–437. http://www.columbia.edu/~en2198/papers/retail_wholesale.pdf
(accessed May 25, 2010).
Nakamura, E., and J. Steinsson. 2008. “Five Facts about Prices: A Reevaluation of Menu Cost
Models.” Quarterly Journal of Economics 123(4): 1415–1464.
http://www.columbia.edu/~en2198/papers/fivefacts.pdf (accessed May 25, 2010).
———. 2009. “Lost in Transit: Product Replacement Bias and Pricing to Market.” Working
Paper. New York: Columbia University (accessed May 25, 2010).
http://www.columbia.edu/~en2198/papers/ippsubs.pdf
Norwood, J., C. Carson, M. Deese, N.J. Johnson, F.S. Reeder, J.E. Rolph, and S. Schwab.
2006.OffShoring: An Elusive Phenomenon. Report of the Panel of the National Academy
of Public Administration prepared for the U.S. Congress and the Bureau of Economic
Analysis, Washington, DC.
Parker, R.P. 2004. “BEA Improves Consistency and Timeliness of Industry Accounts.” Business
Economics, October.
Reinsdorf, M. 1993. “The Effect of Outlet Price Differentials in the U.S. Consumer Price Index.”
In Price Measurements and Their Uses, M.F. Foss, M.E. Manser, and A.H. Young, eds.
NBER Studies in Income and Wealth. 57: 227–254.
———. 1994a. “The Effect of Price Dispersion on Cost of Living Indexes,” International
Economic Review 35: 137–149.
———. 1994b. “New Evidence on the Relation between Inflation and Price Dispersion.”
American Economic Review 84: 720–731.
Stigler, G., ed. 1961. The Price Statistics of the Federal Government. Report to the Office of
Statistical Standards, Bureau of the Budget. Cambridge, MA: National Bureau of
Economic Research.
Strassner, E.H., R.E. Yuskavage, and J. Lee. 2009. “Imported Inputs and Industry Contributions
to Economic Growth: An Assessment of Alternative Approaches.” Presented at the
conference “Measurement Issues Arising from the Growth of Globalization,” held in
Washington, DC, November 6–7.
264
Streitwieser, M.L. 2009.“A Primer on BEA’s Industry Accounts.” Washington, DC: U.S. Bureau
of Economic Analysis.
http://www.bea.gov/scb/pdf/2009/06%20June/0609_indyaccts_primer.pdf
Uchitelle, L. 2009. “Economists Seek to Fix a Defect in Data That Overstates the Nation’s
Vigor.” New York Times, November 8.
Walsh, C.M. 1901. The Measurement of General Exchange Value, New York: Macmillan and
Co.
———. 1921. The Problem of Estimation. London: P.S. King & Son.
Yuskavage, R.E., E.H. Strassner, and G.W. Medeiros. 2008. “Outsourcing and Imported Inputs
in the U.S. Economy: Insights from the Integrated Economic Accounts.” Paper presented
at the 2008 World Congress on National Accounts and Economic Performance Measures
for Nations, held in Arlington, VA, May 12–17.
265
Are there Unmeasured Declines in Prices of Imported
Final Consumption Goods?
Marshall Reinsdorf and Robert Yuskavage*
U.S. Bureau of Economic Analysis
Written for Presentation at the
conference “Measurement Issues Arising from the Growth of Globalization”
Sponsored by the
and
November 6–7, 2009
Washington, DC
____________
* The views in this paper are those of the authors and should not be attributed to the Bureau of Economic
Analysis. We are grateful for the assistance of Brian Lindberg. Authors’ emails:
[email protected] and [email protected].
267
One of the ways in which the U.S. economy has been transformed since the 1970s is the growth
of its engagement in international trade. Imports of nonpetroleum goods have risen particularly
rapidly. They amounted to about 10 percent of final domestic uses of goods in personal
consumption expenditures and gross investment in 1975, about 20 percent in 1991–1992, and
about 30 percent in 2008 (Figure 1).
Many factors have been found to contribute to growth of imports, including lower transport
costs, lower communications costs, advances in managing the logistics of fragmented supply
chains, multilateral trade liberalizations, scale economies, growth in varieties, rising productivity
and capital stocks of emerging Asian economies, and undervalued foreign currencies. These
factors can be expected to act to lower prices of imports to domestic buyers. Substitution to
lower-priced sources of supply therefore seems to be an important driver of the growth of
nonpetroleum goods imports.
Yet the U.S. import price indexes generally give no indication that prices have been a significant
factor in the displacement by imports of domestic sources of supply. Goods prices at the
consumer level should be an average of prices from domestic and imported sources of supply
adjusted for trade and transportation margins, so if import prices are falling relative to domestic
prices, we might expect to see the price index for imports of nonpetroleum goods rising more
slowly (or falling faster) than a price index for nonenergy goods in personal consumption
expenditures (PCE) in the national income and product accounts (NIPAs). Although this pattern
did occur between 1996 and 2001, in most years since 1990 the imports index has risen faster
than an index of PCE nonenergy goods prices (Figure 2). On a cumulative basis over the time
span from 1990 to 2008, prices for imports of nonpetroleum consumer goods rose about 4
percent more than the PCE prices.
Figure 2 suggests that the import indexes might be missing some price declines, but comparisons
of aggregate indexes like those shown in Figure 2 can be affected by differences in commodity
composition or weights. If commodities with above-average rates of price growth are more
heavily represented in the mix of items included in the nonpetroleum goods import index,
declines in relative prices of imports might have occurred at the level of detailed items but have
been masked in the more aggregated indexes by the mix effects.
TWO HYPOTHESES CONCERNING THE BEHAVIOR OF THE IMPORT PRICE
INDEX
Newly Imported Products
Imports could gain market share through lower prices in two ways that would be difficult for the
import price indexes to measure. First, when substitution from a domestic supplier to a foreign
supplier involves a product that was not previously imported, that product constitutes a new good
for purposes of the import price index. This means that it will be linked into the import price
index in a way that avoids an effect on the index’s level. If the newly imported product starts out
priced on a par with competing domestic products (perhaps subsequently picking up market
share by falling in relative price), and it is brought into the import index promptly, the index will
269
not suffer any bias. (Of course, bringing the new import into the index promptly may not be
easy, particularly if it does not fit into the preexisting product classification system used for the
imports index.) Yet a more likely scenario for a newly imported product that succeeds in
capturing a significant market share is that it already has a substantially lower quality-adjusted
price than the domestic incumbents at the time of entry. If an imported product enters at a low
price, a temporary state of disequilibrium may ensue as information about the price and quality
of the newly available import diffuses and preexisting contracts with incumbent suppliers expire.
When entry occurs at a substantially lower price level than that prevailing in the marketplace,
prompt linking of the new import into the buyers’ price index is not sufficient to capture all the
gains to buyers. A theoretical measure all of the gains would bring the newly imported product
into the index by creating a hypothetical observation for the time period before the one in which
actual purchases of the product begin with a quantity of zero and a price equal to the Hicksian
virtual price (which is defined as the price that is just high enough to drive demand to zero). The
consumer surplus from the drop from the Hicksian virtual price to the price at which purchases
are first occur can then be measured by integrating the area under a compensated demand curve.
A practical approximation to this theoretical concept would use a Törnqvist, Fisher, or Fisherlike index formula and an estimated value of the virtual price at which quantity demanded would
be zero. For example, if the entrant offers a quality level that is identical to that of the incumbent
supplier, the Hicksian virtual price equals the price of the incumbent supplier. This principle
was used by Griliches and Cockburn (1995) to argue that when a branded pharmaceutical goes
off patent, the low-priced generic should not be linked into the CPI. Instead, the prices of the
generic and its branded counterpart should be directly compared, with a quality adjustment for
the generic that attributes half of the savings enjoyed by those consumers who substitute to the
generic to a quality decline and counts the other half as a pure price reduction. Similarly,
Reinsdorf (1993) argues that when consumers change their purchasing patterns to lower-priced
discount outlets, linking the lower-priced outlets into the CPI would result in outlet substitution
bias. As in the case of generic pharmaceuticals, the Hicksian virtual price for those consumers
who substitute the discount outlet for the full-service outlet can be estimated by the price of the
full-service outlet.
A buyers’ price index that combined imported and domestic sources of supply could also
properly account for the gains from the entry of a new product into the imports basket if the
newly imported product matched the characteristics of a domestically supplied product. In the
case of matching characteristics, the import could be brought into the buyers’ price index as a
directly comparable substitute, meaning that its price would be directly compared with the price
of the domestically supplied item that it replaced. In addition, if the characteristics are not
exactly identical, a direct quality adjustment could be made to the price of the imported item to
allow it to be compared to its domestically supplied counterpart. In many cases, imported
sources of supply will involve extra delivery lags, communications problems, and warehousing
costs, so that even if the physical characteristics are identical to those of the domestic product
that it replaces, some downward quality adjustment is warranted. In a buyers’ price index, it
might therefore be reasonable to treat a physically identical import the same way that a generic
drug is treated in the CPI index for pharmaceuticals of Griliches and Cockburn (1995).
270
New Product Varieties
The second hard-to-measure way that imports could gain market share is through the entry of
new varieties (or source countries) of products that are already present in the imports index’s
basket. A difficult challenge in the construction of import (and export) price indexes is that price
changes tend to coincide with changes in product specifications. (This may be because foreign
trade occurs under contracts that fix the price for the lifespan of the variety, and the signing of a
new contract presents an opportunity to revise both the price and the item characteristics.) The
effect on the price of a change in an item’s characteristics is difficult to measure directly, so new
varieties are usually linked in a way that avoids any effect on the index’s level. This practice
tends to result in an index that is artificially flat if the market is one in which price changes tend
to be delayed until the item’s characteristics are updated.
In contrast to the “new goods” hypothesis, which always implies an upward bias, the new
varieties hypothesis implies that an upward bias in the import prices indexes arises only for items
whose true price trend (including correctly measured quality change effects) is downward. Many
electronic and technology-intensive goods do, in fact, have falling price trends, so a bias toward
zero in the measured rate of price change could plausibly have a positive effect on the rate of
growth of the aggregate import index for nonpetroleum goods.
Nakamura and Steinsson (2008) find that in the samples used to construct the imports indexes,
prices frequently remain constant for the life of a quote (that is, for the length of time that the
specific version of an item remains in the sample). Respondents tend to report a price change
only at the time of a change in the version of the item that they import. When price and
characteristics change simultaneously, separating the reported change in price into a quality
change component and a pure price change component is difficult. Consequently, the new
version of the product is typically linked into the index, which is equivalent to attributing all of
the reported price change to quality change. The fraction of the observed price changes that are
effectively treated as price changes of zero is high for some products in the import and export
indexes. This is not the case of the CPI, because most products in the CPI do not behave in this
way, and because the CPI uses hedonics or other methods to value characteristics changes for
those products where price changes tend to coincide with characteristics changes.
USE OF CPIs FOR COMPARISONS
Construction of Suppliers’ Price Indexes and Purchasers’ Price Index
Although no buyers’ index that allows price comparisons of imported and domestic-sourced
products exists at the wholesale level or for intermediate inputs, buyers’ indexes for final
consumer goods do exist. In particular, CPIs for individual goods or classes of goods are buyers’
indexes that can include both imported and domestically produced items.
Although the CPI does bring in some foreign-sourced items by linking, it appears to be less
vulnerable to linking bias than the import price index. In the CPI, a foreign-sourced item would
271
be directly compared to a domestic-sourced item if the consumer is thought to perceive their
quality levels as the same. Furthermore, the CPI makes more use of direct quality adjustment
techniques, such as hedonic regressions and suppliers’ cost estimates, than do the import and
export indexes.
On the assumption that some declines in prices paid by buyers from substitutions from domestic
to imported versions of products and from substitutions between imported versions of products
are reflected in the CPI but not in the imports price indexes, CPIs should tend to show lower
inflation rates than import price indexes covering similar detailed items. In Figure 3, for
example, the average rate of growth of tire prices is 4.1 percent in the import price indexes, but
only 3.2 percent in the CPI. Moreover, if the growth rate of the CPI for tires is an average of true
import price growth rate and the correctly measured growth rate of the PPI, the lower rate of
growth of the CPI than of the PPI implies that true growth rate of the import prices must be
below that of the CPI.
To take a more systematic approach to comparing import prices and CPIs for detailed items, we
used BEA’s industry accounts data to identify commodities included in PCE that are at least
partly supplied by imports. The industry accounts also show the proportion of total domestic
supply of each commodity that come from imports. Using this weighting information we
constructed Fisher indexes that combine prices received by domestic producers (which BEA
measures based on PPIs from BLS) and prices received by suppliers of imports (which BEA
measures based on import price indexes from BLS). Even “producer prices” can include imports
in BEA’s industry accounts, at BLS “producer price indexes” do not include imports, so we term
these indexes suppliers’ price indexes.
Prices at the retail level include transportation margins, wholesale and retail distribution margins,
and commodity taxes. Therefore, in addition to our indexes of suppliers’ prices, we constructed
purchasers’ price indexes that combine the suppliers prices with price indexes for transportation
and distribution margins and adjust for changes in commodity taxes. Our purchasers’ price
indexes represent predicted CPIs based on prices and weights from BEA’s industry accounts.
However, we are not entirely confident of the quality of some of the price indexes for
distribution margins, so we will compare both our suppliers’ price index and our purchasers’
price index with a CPI at the most detailed level of aggregation that is available.
If we take the CPI and the supplier price index as given, the equation that expresses the logchange in the CPI as a weighted average of log-changes in the supplier price index and the price
index for transportation and distribution margins and taxes contains only one unknown value,
that of the price index for distribution margins and taxes. We can therefore solve this equation
for the implied price index for the margin industry services and taxes. Assuming that the CPI is
correct and that the quality of the match between the CPI and the suppliers’ price index is good,
an implausible value for this implied index would imply that the suppliers’ index is biased. If we
also assume that PPI is correct, then the bias in the suppliers’ price index would have to come
from its imports component.
Under the assumption that the prices of inputs into transportation and distribution are not
changing and that tax rates are not changing, the rate of decline in the implied price index for the
272
margin industries will equal their rate of productivity growth. If the implied productivity growth
rate in transportation and distribution is implausibly high, that is evidence of either upward bias
in the suppliers price index, downward bias in the CPI, or mismatch between the microlevel
composition of the detailed CPI that we used and the microlevel composition of our suppliers
index.
Aggregation as a Solution to the Problem of Poor Match Quality
At the level of detailed comparisons, the quality of the matches between CPIs and suppliers’
price indexes is often dubious. However, if poor match quality is a source of random noise in the
comparisons that is as likely to be positive as negative, a consistent pattern of implausibly high
implied estimates of productivity growth in transportation and distributions would still suggest a
systematic downward bias in the import indexes. In addition, the industry accounts data show
the importance of each commodity in final uses in personal consumption expenditures (PCE),
allowing us to use appropriate weights to aggregate commodities. Problems of misclassification
become much less important at aggregate levels such as all durable goods, all nondurable goods,
and all goods that have imported sources of supply. Therefore, comparisons made at aggregate
levels are likely to be robust problems with matches between CPI items and our detailed
suppliers’ price indexes.
The main impediment to constructing good matches between CPIs and our suppliers’ price index
is that the most detailed CPIs available are generally broader in coverage than the commodity
categories in the industry accounts. For example, fur coats are a commodity in our industry
accounts data, but BLS does not publish a CPI for fur coats. We therefore had to match fur coats
to a CPI for women’s coats in general. To give another example, we matched boat building in the
industry accounts to a CPI for recreational vehicles including bicycles. The unavailability of
sufficiently detailed CPIs means that at the level of individual items, many of the comparisons of
CPIs to our suppliers and purchasers indexes do not hold the commodity mix constant. This
problem becomes less severe when detailed items in the industry accounts are aggregated.
Two other caveats are also worth noting. First, the indexes that we use have positive variances,
which we have not attempted to estimate. Second, the import price indexes exclude tariffs, but
tariffs undoubtedly influence the retail prices for imported items that are measured by the CPI.
Tariff rates have trended down in recent decades, so declines in tariffs have probably acted to
reduce the growth rate of the CPIs compared to those of the import price indexes. Nevertheless,
for most items with significant volumes of imports, average effective tariff rates started out at
low levels, so most of the reductions during the period covered by our sample were modest.
EMPIRICAL RESULTS
Suppliers’ Price Index Comparisons
Differences in average annual growth rates of suppliers’ indexes from matched CPIs between
1997 and 2007 are shown in the first column of Table 1. For most nondurables types of goods,
273
the indexes of suppliers’ prices do not differ significantly from their CPI counterparts. Indeed, in
the cases of food and alcoholic beverages, the difference in growth rates is zero after rounding.
For the other categories of goods, however, the suppliers’ indexes rise more than their CPI
counterparts. In the case of apparel and textiles, the growth rate of the suppliers’ price index
exceeds that of the CPI by 1.5 percent per year. At the retail level, seasonal apparel items tend to
enter at a high price at the beginning of the selling season, then go on sale later in the selling
season before they exit the marketplace. This pattern of falling prices during the life of a quote
would cause downward bias in the CPI if at the beginning of a selling season quotes tend to enter
the index via linking rather than via a comparison with the final price of the previous selling
season. However this problem appeared to have been largely resolved in 1991 when hedonic
methods were introduced into the CPI for apparel.
Another factor that likely contributes to the gap between the apparel indexes’ growth rates is the
use of different index formulas to construct elementary aggregates. A geometric mean index
formula is used to construct most of the elementary aggregates in the CPI, including apparel, but
a Laspyeres-like formula is used in the IPP and the PPI. Geometric mean indexes have a number
of desirable axiomatic properties that Laspyeres-like indexes lack, and under certain assumptions
they do a better job of accounting for substitution behavior. The dataset used in Feenstra et al.
(2009) shows that the effect of changing to a Törnqvist formula (which resembles a geometric
mean formula) on the import price indexes would reduce the growth rate of the apparel indexes
by about 0.3 percent per year. On the other hand, comparisons of the CPI-Uto the CPI-RS from
1991 to 1998—years when the CPI-U did not use geometric means—imply that the effect of
adopting the geometric mean formula on the CPI for apparel is about 1.3 percent per year. Such
a large effect is troubling, as it suggests that quotes linked out of the index at sale prices may
have a disproportionate effect on the geometric mean formula. Thus, downward bias in the CPI
for apparel could account for some of the gap between its growth rate and that of the suppliers’
index. The gap between the suppliers’ index and the matched CPI for apparel is not large
enough to constitute evidence of the existence of an upward bias in the imports price index
considering the possible presence of downward bias in the CPI and the noisiness in the index
comparisons arising from differences in index composition.
On the other hand, for durable goods other than motor vehicles, the growth rate gaps are larger
than for apparel. The most troubling growth rate discrepancies are for computers and related
equipment, where the suppliers’ price index grows 6.4 percent per year faster than the CPI, and
other electronic equipment, where it grows 4.2 percent per year faster. However, even for
durables in general, the growth rate gap is pretty high, at almost 2 percent per year.
Purchaser Price Index Comparisons
Adding in distribution and transportation margins to obtain purchasers’ price indexes makes the
growth rate gaps smaller for nondurable and apparel, reducing the overall average gap from 1
percent per year to 0.6 percent per year, or from 1.1 to 0.7 percent per year if tobacco is
excluded. (Note however, that to be in our sample, an item in the industry accounts needed to
have positive imports, few or no missing values, and a CPI counterpart, so “all items” in our
sample covers only about 20 percent of personal consumption expenditures on nonenergy
274
goods.) Indeed, a number of nondurable goods have negative growth rate differences between
their purchasers’ price index and the corresponding CPI.
On the other hand, the growth rate differentials for most apparel items still remain above 1
percent per year. What is more, the product categories with large gaps between their CPIs and
their suppliers’ indexes have even larger growth rate gaps using a purchasers’ index. In
particular, the growth rate gap for electrical equipment excluding computers rises to 4.7 percent
per year, while the growth rate gap for computers and related equipment rises to 11.6 percent per
year.
The electronic and computer items have falling CPIs, so the upward bias in their import indexes
is consistent with Nakamura and Steinsson’s (2009) theory that a pattern of price changes
coinciding with linking causes excessive flatness in the import indexes.143 Furthermore, in the
case of tobacco products, the negative gaps of the suppliers’, purchasers’, and import price
indexes compared with the CPI are also consistent with the hypothesis of artificial flatness in the
imports index, as these products have high rates of price growth in the CPI.
Similarity of PPIs and Import Price Indexes for Certain Items
Simple comparisons of import price indexes with corresponding aggregates of domestic prices as
measured by producer price indexes can also shed light on relative growth rates of the import
indexes. In constructing the aggregates used for these comparisons, we weight the indexes for
detailed commodities in proportion to the importance of these commodities in final goods uses as
measured in the industry accounts.
These comparisons also suggest the presence of an upward bias in the import index for durable
goods, but not in other categories of goods. In the case of nondurable goods, domestic prices at
the producer level rise faster than prices at the consumer level, while import prices rise more
slowly than consumer prices. Thus, the closeness of the suppliers’ indexes to the CPIs is a result
of offsetting effects of slow growth in import prices and fast growth in producer prices. For
apparel and textile items, import and domestic producer prices both differ from CPI growth rates
by about +1.5 per cent per year. However, for durable goods, the growth rate of import indexes
is 2.3 percentage points above that of the corresponding CPI and 0.7 (= 2.3 – 1.6) percentage
points above that of domestic producer prices. For computers and peripherals, domestic
producer prices fall nearly as fast as the CPI, but the growth rate of import index is about 11.8
percent per year above that of the rapidly falling CPI.
Implied Productivity Growth in Transportation and Wholesale and Retail Distribution
Under neoclassical assumptions, the difference between the growth rate of the price index for the
output of an industry and the price index for the inputs that it uses is an estimate of its
productivity growth. Price indexes for labor and other inputs are unlikely to have growth rates
143
From January 1998 to September 2003, CPI computer price indexes used hedonic regressions for quality
adjustment, and since then these indexes have used direct methods to estimate attribute values. These techniques
allow measurement of price changes that are time to coincide with changes in attributes.
275
below zero, so reversing the sign of the growth rate of the implied price index for transportation
and distribution services gives a lower bound estimate of productivity growth in these services.
Solving for the price index for transportation and distribution services that would explain the
difference between our suppliers price index and the matched CPI, we find a plausible positive
rate of growth for nondurables other than apparel of 1.3 percent per year and a not inconceivable
growth rate for apparel and textile products of -2.5 percent per year (Table 2). On the other
hand, for durable goods the implied price change for transportation and distribution is about -8
percent per year, which is too low to be believed. For computers and peripherals, the implied
growth rate is almost -30 percent per year. Although strong productivity growth in distribution
services is plausible, rates as high as 8 percent per year or more are not plausible. They therefore
suggest that the difference between the growth rates of the import and domestic producer price
indexes and the growth rate of the CPI is too large to be correct.
RELATION BETWEEN IMPORTS AND WHOLESALE AND RETAIL DISTRIBUTION
MARGINS
Price reductions that are realized by substituting foreign sources of supply for final consumption
items for domestic ones are unlikely to be completely passed on to consumers. Instead, some of
these price reductions are likely to result in expansions of margins received by the wholesale and
retail distribution sectors. One reason for this is that more distribution services are required to
set up and manage international supply chains. In addition, distributors are likely to have higher
inventory costs and greater risks of being stuck with unwanted inventory when suppliers are
distant and turnaround times for restocking are long. In addition, while the process of switching
to foreign sources of supply is under way, markets are likely to be in a temporary disequilibrium
that allows early switchers to earn economic rents.
To test whether higher proportions of imports in the overall domestic supply of a commodity are
associated with higher distribution margins, we regress trade margin levels and growth rates on
import share levels and growth rates. The regression in Table 3 implies that a 10 percent
increase in the share of domestic supply sourced from imports is associated with a 1.3 percentage
point expansion in the distribution margin.
Commodities that are heavily imported—such as apparel—might also have characteristics that
require lots of distribution services. If so, import share could be a proxy for the types of
characteristics that make a commodity require more distribution services, resulting in upward
bias in the regression coefficient in Table 3. We therefore also test for a relationship between the
growth of imports as a share of total commodity supply and the growth of distribution margins.
The growth rate regression also shows a positive and statistically significant relationship between
imports and distribution margins (Table 4). The regression coefficient implies that a commodity
with a 10 percentage point increase in its import share would have 0.93 percentage points more
growth in its margin rate than a commodity with no change in its import share. Thus, the
theoretical prediction of a link between imports and margins received by the distribution
industries finds some empirical support.
276
CONCLUSION
The increased international engagement of the U.S. economy has enhanced the roles of the
import and export price indexes in the measurement of real output growth. For imported final
goods that enter personal consumption expenditures in the NIPAs, an upward bias in the import
index would result in upward bias in the measure of GDP; these items are deflated by an import
index in the M component of the formula GDP = C + I + G + X – M but then deflated by a
consumer price index when they reach the C component.
In the case of nondurable goods, comparisons of import indexes and of suppliers’ and
purchasers’ price indexes with CPIs at the most detailed level possible given the available data
indicate no systematic differences in behavior of these indexes. But for apparel and textile
items, the import prices seem to grow faster than CPIs by about 1.5 percent per year, and for
durable goods they seem to grow faster by more than 2 percent year. A very large discrepancy
of over 11 percent per year for computers contributes significantly to this discrepancy for durable
goods. Furthermore, the implied rates of change in prices for transportation and distribution
services for durable goods are implausibly negative. The index comparisons therefore suggest
the presence of a significant upward bias in the import price indexes for some types of durable
goods, especially computers, as well as the possible presence of a modest upward bias in the
import indexes for apparel. Nevertheless, these comparisons are not definitive evidence of the
existence of a problem because the CPIs used in the comparisons often differ in their detailed
item composition from the indexes with which they are compared.
The results in this paper are consistent with Nakamura and Steinsson’s finding of a bias toward
zero in the rate of growth of the import price indexes caused by linking out of a large fraction of
the price changes that occur. The substantial positive discrepancies between the growth of the
import index and the growth of the CPIs occur for the apparel and durable goods items that have
falling CPIs, and the largest negative discrepancy occurs for an item with a very high growth rate
in the CPI, tobacco products. Further research on solutions to the linking problem identified by
Nakamura and Steinsson could yield important benefits for our measures of import and export
prices, and improve our measures of the growth of output and productivity, especially at the
industry level.
Finally, we note that if any bias that is present in the import indexes is matched by a similar bias
in the export indexes if exports and imports had the same nominal value, the net effect of the two
biases on our estimates of real GDP growth would be zero. However, complete cancellation is
unlikely to occur in practice because exports of goods are less than imports of nonpetroleum
goods (the difference is between 2 and 4 percent of GDP in most of the years in our sample) and
because some sources of upward bias in the import indexes are either not present or less
important on the export side. In particular, imported versions of many products are likely to
have entered the U.S. market at a significantly lower price level than the incumbent domestic
version of the product. This results in a gain to consumers that should theoretically be accounted
for by use of a high Hicksian virtual price in period before entry was observed to bring the
imported version into the import index. Yet on the export side, when new opportunities to export
a particular product arise, the Hicksian virtual price would be lower than the first observed
277
selling price, implying the presence of a downward bias in the export index. Alternatively, we
could model new exports as having been caused by positive technology shocks to exporters and
ask how much lower the first observed price is than the virtual price at which exports would first
become profitable with technology held constant at the new level. This would yield a correction
of the same sign as the one that is applicable to the import index. Yet growth of exports has been
less than growth of imports, and the value of the export products that could reasonably be
modeled in this way is undoubtedly less than the value of the new types of imports.
278
References
Moses, Karin, and Brent Moulton. 1997. “Addressing the Quality Change Issue in the
Consumer Price Index.” Brookings Papers on Economic Activity 1: 305–349.
Feenstra, Robert C., Benjamin R. Mandel, Marshall B. Reinsdorf, and Matthew J.
Slaughter. 2009. “Effects of Terms of Trade Gains and Tariff Changes on the
Measurement of U.S. Productivity Growth.” NBER Working Paper 15592.
Nakamura, Emi, and Jón Steinsson. 2009. “Lost in Transit: Product Replacement Bias
and Pricing to Market.” NBER Working Paper 15592. Cambridge, MA: National
Bureau of Economic Research.
279
Figure 1: Nonpetroleum Goods Imports as Percent of GDP, PCE
Goods and PCE Goods+Investment
50
45
GDP
40
PCE Goods and Gross Investment
PCE Goods
35
30
25
20
15
10
5
0
1975
1980
1985
1990
1995
2000
2005
Figure 2: Price Indexes for Consumer Good Imports and for
Personal Consumption Expenditure Goods ex Energy
114
112
110
108
106
PCE goods excluding energy
104
Imports of Consumer Goods (including automotive vehicles)
102
100
1990
1992
1994
1996
1998
280
2000
2002
2004
2006
2008
281
Table 1: Growth Rate Differences from Matched CPIs of Suppliers and Purchasers Price Industry,1997-2007
-0.6
0.1
-0.5
-0.2
-0.4
-6.6
Domestic
Prices
0.6
0.0
0.0
1.3
1.1
-0.5
MEMO:
Index of
Matched
CPIs
2.2
2.1
1.9
1.5
1.4
8.1
Difference from Matched CPI
0.3
0.0
0.0
0.6
1.1
Purchasers’
prices
-0.6
-0.7
-0.6
-0.1
0.2
-3.3
Durables
Motor vehicles and parts
New cars and trucks
Electrical equipment ex. computers
Computers, peripherals and software
Furniture and wood products
Clocks and watches
Tools, hardware and supplies
Other durables
1.9
0.2
0.4
4.2
6.4
2.3
1.8
1.8
3.0
2.0
0.2
0.5
4.8
11.7
1.4
1.7
0.9
1.9
2.3
0.7
1.2
3.5
11.8
1.5
1.8
1.7
3.1
1.6
-0.1
-0.2
4.3
3.8
2.5
1.9
1.7
2.4
-2.2
-0.1
-0.6
-5.6
-20.8
-0.6
-1.4
-0.2
-0.8
Apparel and textiles
Women's and girls' apparel
Men's and boy's apparel
Other apparel
Footwear
Textile and sewing products
1.5
1.9
1.3
2.4
0.6
1.5
1.4
1.7
1.4
1.7
0.5
1.1
1.5
1.9
1.4
2.4
0.6
1.4
1.5
1.8
0.7
2.4
1.2
1.6
-1.2
-1.5
-1.5
-1.2
-0.4
-0.8
All products
All products ex tobacco
1.0
1.1
0.6
0.7
0.7
1.0
1.1
1.1
0.2
-0.1
Suppliers’ prices
Nondurables ex apparel
Food
Alcohol
Misc. household supplies
Paper products, books and magazines
Tobacco products
-0.6
282
Import Prices
Table 2: Growth Rates of Price Index for Transportation and Distribution Services implied by Difference between Suppliers Price
Index and Matched CPI, 1997-2007
Implied price index for Transport &
Distribution
Nondurables ex. apparel
Food
Alcohol
Misc. household supplies
Paper products, books and magazines
Tobacco products
Durables
Motor vehicles and parts
New cars and trucks
Electrical and electronic equipment ex.
Computers
Computers, peripherals and software
Furniture and wood products
Clocks and watches
Tools, hardware and supplies
Other durables
Apparel and textiles
Women's and girls' apparel
Men's and boy's apparel
Other apparel
Footwear
Textile and sewing products
All products
All products ex tobacco
Actual price index for Transport &
Distribution
1.3
0.5
2.2
1.9
1.1
0.3
8.7
-7.9
-1.3
-1.8
-11.0
0.2
0.9
0.3
0.4
2.4
0.1
0.3
0.4
0.2
-29.7
-2.7
-1.7
-1.8
-3.3
-2.5
-3.0
-2.6
-3.0
-0.9
-2.2
-0.8
-1.2
283
0.1
0.0
0.1
0.0
0.1
-0.0
-0.0
-0.0
0.0
0.0
0.0
0.3
0.2
Table 3: Regression of Average Level of Distribution Margin on Share of Domestic Supply from Imports
Coefficient
t statistic
Intercept
0.3663
29.8
Share supplied by imports
0.1290
4.3
Growth of share of imports
0.0985
1.4
Table 4: Regression of Growth of Distribution Margin from 1997 to 2006 on Share of Domestic Supply from Imports
Coefficient
t statistic
Intercept
0.0067
1.2
Share supplied by imports
0.0272
1.9
Growth of share of imports
0.0934
2.8
284
SESSION 6: MEASUREMENT OF IMPORT AND INPUT
PRICES
CHAIR: Julia Lane (National Science Foundation)
Offshoring and the State of American Manufacturing, Susan Houseman (Upjohn Institute),
Christopher Kurz (Federal Reserve Board), Paul Lengermann (Federal Reserve Board), and
Benjamin Mandel (Federal Reserve Board)
Producing an Input Price Index, William Alterman (Bureau of Labor Statistics)
DISCUSSANT: Emi Nakamura (Columbia University)
285
Offshoring and the State of American Manufacturing
Susan Houseman+
Christopher Kurz*
Paul Lengermann*
Benjamin Mandel*
June 2010
ABSTRACT
The rapid growth of offshoring has sparked a contentious debate over its impact on the U.S.
manufacturing sector, which has recorded steep employment declines yet strong output growth—
a fact reconciled by the notable gains in manufacturing productivity. We maintain, however, that
the dramatic acceleration of imports from developing countries has imparted a significant bias to
the official statistics. In particular, the price declines associated with the shift to low-cost foreign
suppliers generally are not captured in input cost and import price indexes. To assess the
implications of offshoring bias for manufacturing productivity and value added, we implement
the bias correction developed by Diewert and Nakamura (2009) to the input price index in a
growth accounting framework, using a variety of assumptions about the magnitude of the
discounts from offshoring. We find that from 1997 to 2007 average annual multifactor
productivity growth in manufacturing was overstated by 0.1 to 0.2 percentage point and real
value added growth by 0.2 to 0.5 percentage point. Furthermore, although the bias from
offshoring represents a relatively small share of real value added growth in the computer and
electronic products industry, it may have accounted for a fifth to a half of the growth in real
value added in the rest of manufacturing.
287
JEL Classification Codes: O41, O47, F14, L60
Key Words: offshoring, manufacturing, price measurement, productivity, output growth
Acknowledgments: This research was supported with funding from the Bureau of Economic
Analysis and the Alfred P. Sloan Foundation, and was conducted with restricted access to Bureau
of Labor Statistics (BLS) import price data as well as unpublished detail on imported materials
provided to us by the Bureau of Economic Analysis (BEA). The analysis and conclusions set
forth are those of the authors and do not indicate concurrence by other members of the research
staff of the Board of Governors, the BLS, or the BEA. We thank Jonathan Collins and Lillian
Vesic-Petrovic for research assistance and participants of the conference “Measurement Issues
Arising from the Growth of Globalization,” held in Washington, DC, November 2009, for
comments and suggestions. We are especially grateful to Bill Alterman, Erwin Diewert, Nicole
Mayerhauser, Emi Nakamura, Alice Nakamura, Marshall Reinsdorf, Erich Strassner, and Rozi
Ulics for their invaluable feedback and assistance on this project.
+Houseman ([email protected]) Upjohn Institute for Employment Research
*Kurz ([email protected]); Lengermann ([email protected]),
Mandel ([email protected]) Federal Reserve Board
288
Offshoring and the State of American Manufacturing
Developing economies have become the new, low-cost suppliers of a wide range of products
purchased by consumers and used as intermediate inputs by producers, with China—now the
largest exporter to the United States—accounting for about a third of the growth in commodity
imports over the last decade. The rapid growth of offshoring—defined as the substitution of
imported for domestically produced goods and services—contributed to a ballooning trade
deficit and sparked a contentious debate over its impact on the U.S. manufacturing sector, which
shed 20 percent of its employment, or roughly 3.5 million jobs, from 1997 to 2007. Concerns
over employment losses and the trade deficit have prompted a recent spate of government and
private sector proposals to revitalize manufacturing.144
Our paper highlights the dramatic growth of offshoring and the structural changes occurring in
manufacturing in the decade prior to the current recession. During this time, more than 40
percent of imported manufactured goods were used as intermediate inputs, primarily by domestic
manufacturers. Using a growth accounting framework, we examine the contributions to the
growth in real (constant price) domestic shipments in manufacturing from the inputs to
production and from multifactor productivity (MFP). A novel feature of our analysis is that we
distinguish between imported and domestic materials inputs, which enables us to more closely
examine offshoring by manufacturers. We find substantial evidence of offshoring. The
contribution from imported materials to the growth in real manufacturing shipments was larger
than that of any other factor input and was more than twice the contribution from capital. At the
same time, contributions from domestic materials and, reflecting declining employment, labor
were negative.
In spite of the steep employment declines in manufacturing, official statistics indicate that real
value-added growth in U.S. manufacturing was robust, increasing almost as quickly as that for
all nonfarm business. What happened in manufacturing? As put by former Labor Secretary
Robert Reich (2009), “In two words, productivity growth.” Indeed, the disparate trends in
manufacturing—steep employment declines and strong output growth—are reconciled in the
data by high productivity growth. While Reich and others have cited the strong output and
productivity figures as evidence of the strength of the American manufacturing, we discuss
reasons to qualify this conclusion.
First, the robust output and productivity growth in manufacturing is largely attributable to one
industry: computer and electronic products manufacturing. The average annual growth rate of
value added in manufacturing excluding computers—which accounted for about 90 percent of
manufacturing value added throughout the period—was less than a third of the published growth
rate for all manufacturing. As a result, the aggregate numbers do not accurately characterize
trends in most of manufacturing.
Second, the price declines associated with the shift to low-cost foreign suppliers generally are
not captured in price indexes. The problem is analogous to the widely discussed problem of
144
See, for example, Executive Office of the President (2009), Helper (2008), New America Foundation (2010),
Pisano and Shih (2009), Pollin and Baker (2010), and Surdna Foundation (2010).
289
outlet substitution bias in the literature on the Consumer Price Index (CPI). Just as the CPI fails
to capture lower prices for consumers due to the entry and expansion of big-box retailers like
Wal-Mart, import price indexes and the intermediate input price indexes based on them do not
capture the price drops associated with a shift to new low-cost suppliers in China and other
developing countries. A bias to the input price index from offshoring implies that the real growth
of imported inputs has been understated. And if input growth is understated, it follows that the
growth in MFP and real value added have been overstated.
Building upon Diewert’s (1998) characterization of the bias from outlet substitution to the CPI,
Diewert and Nakamura (2010) demonstrate that the bias to the input price index is proportional
to the growth in share captured by the low-cost supplier and the percentage discount offered by
the low-cost supplier. Although the actual input price changes from offshoring are not
systematically observed, evidence from import price microdata from the Bureau of Labor
Statistics (BLS), industry case studies, and the business press indicate that there are sizable
discounts from offshoring to low-wage countries.
If this evidence is representative of the actual discounts manufacturers realized from offshoring,
then the biases to MFP and value-added growth may well be significant. We estimate that
average annual MFP growth in manufacturing was overstated by 0.1–0.2 percentage points and
real value-added growth by 0.2–0.5 percentage points from 1997 to 2007. And, although the bias
from offshoring represents a relatively small share of real value-added growth in the computer
industry, it may have accounted for one-fifth to one-half of the growth in real value added in the
rest of manufacturing. Moreover, our work only examines biases to manufacturing statistics from
the offshoring of material inputs. Additional biases may have arisen from the substitution of
imported for domestically produced capital equipment and from the offshoring of services.
These biases have implications not only for the industry statistics, but also for the analyses based
on them. Because the growth of these imports will be understated in real terms, offshoring will,
at least to some degree, manifest itself as mismeasured productivity gains. As a result, studies
that endeavor to assess the impact of low-cost imports on the American economy and its workers
may well understate the true effects.
INTERNATIONAL TRADE AND THE STATE OF AMERICAN MANUFACTURING
One of the most important developments in the U.S. economy in recent years has been the rapid
growth of trade. After being little changed in the early 1990s, the total value of imports and
exports of goods and services jumped from roughly 20 percent of U.S. GDP to 28 percent prior
to the recent downturn. About 80 percent of the increase was attributable to a run up in the value
of imports. The growth of nonoil imports was the most important contributor to the increase in
imports during this period, and nonoil goods imports—largely manufactured goods—accounted
for almost half of total import growth, while oil accounted for about a third and services for the
remainder.145
145
BEA data on the trade in services in 2008 indicate that 59 percent was travel, transport, royalties, and education
related, while the remaining 41 percent was business services.
290
The surge in the imports of manufactured goods—more than 100 percent from 1997 to 2007—
reflects both an increase in the import share of goods for final consumption as well as the import
share of intermediate inputs. According to the BEA, the import share of intermediate material
inputs used by manufacturers increased from under 17 percent in 1997 to 25 percent in 2007. 146
Figure 1 plots this substantial shift in the sourcing of intermediates from domestic to foreign
suppliers.
Also in Figure 1 the imported intermediate materials are classified by type of source country—
developing, intermediate, and advanced. We classify countries with less than 20 percent of U.S.
per capita GDP in 2008 as developing, and with a few exceptions, countries with per capita GDP
equal to or exceeding two-thirds that of that in the United States as advanced. The remaining
countries are classified as intermediate. 147 Developing countries accounted for half of the growth
in foreign materials inputs, with much of that growth coming from China. Intermediate countries,
such as Mexico, accounted for about a third of the growth.
How has the manufacturing sector performed given the growth of imports from low-wage
countries? In particular, has the substantial shift in sourcing “hollowed out” manufacturing or
instead contributed to the emergence of a leaner, more efficient industrial sector? On the one
hand, dramatic drops in employment are often taken to portray a sector in decline. The
precipitous decline in manufacturing employment since the late 1990s is evident in Figure 2 and
is coincident with the rise in foreign sourcing. Employment never rebounded after the 2001–
2002 recession as it had following previous downturns. Indeed, in the decade leading up to the
current recession, manufacturing employment declined by 20 percent, while manufacturing’s
share of employment in the economy fell from 14 percent in 1997 to 10 percent in 2007 (Figure
3). Reflecting plant closures that accompanied the employment declines, the net number of
manufacturing establishments fell by 10 percent from 1998 to 2007. At the same time, the
nominal share of manufacturing value added in GDP fell from 15.4 percent in 1997 to 11.7
percent in 2007.
Statistics on manufacturing production, however, paint a much more favorable picture of the
sector. From 1960 to 2009, the average annual rate of change in real nonfarm business output
was 3.5 percent, only slightly higher than the 3.2 percent average annual change for
manufacturing.148 More recently, from 1997 to 2007, the average annual growth rate of real
manufacturing production was 3 percent, almost the same as the 3.1 percent growth for all
146
Government surveys do not explicitly track the destination of imports to final and intermediate uses. In the data
we employ within this paper, the BEA assumes that an industry’s use of a specific import is in proportion to its
overall use of that commodity in the economy, i.e., the import comparability or proportionality assumption.
147
The main exceptions are Middle Eastern oil producers, which we classify as intermediate countries.
148
Although the average growth of manufacturing has been fairly close to that of the economy as a whole, the
sector has typically exhibited greater cyclical swings. As a result, the sector tends to make outsized contributions to
changes in GDP growth during economic turning points (Corrado and Mattey 1997). In addition, the relatively faster
gains in manufacturing productivity have resulted in lower relative goods prices, which, in combination with
inelastic demand for goods (on average), has led to a decline in manufacturing’s share of nominal output.
291
private industry.149 Moreover, cross-country comparisons show larger production gains in U.S.
manufacturing relative to other advanced industrial countries, according to OECD data.
The divergent trends of employment declines and plant closures, on the one hand, and rapid
growth in real value added, on the other, are primarily reconciled through the lens of
productivity. The steadily increasing series displayed in Figure 3 shows the ratio of output per
hour in manufacturing to output per hour in all nonfarm business since 1960; the series indicates
that labor productivity grew considerably faster in manufacturing throughout the period. Indeed,
the average annual growth rate of labor productivity in manufacturing during 1997 to 2007 was
4.1 percent compared to 2.7 percent for all nonfarm business. Manufacturing labor productivity
also grew substantially faster in the United States than in most other major industrialized
countries during that decade (see BLS 2009). The rapid growth in labor productivity has more
than offset the declines in labor input and has permitted firms to sustain robust growth in real
value added.
Analysts have pointed to the robust output and productivity growth to argue that the
manufacturing sector is relatively healthy.150 Our work, however, suggests the story is more
complex. The aggregate numbers are unrepresentative of the trends in most of manufacturing.
Moreover, we find that the performance of U.S. manufacturing has been overstated to some
extent in the official statistics because of offshoring.
EVIDENCE OF OFFSHORING BY MANUFACTURERS
We begin our analysis by examining the sources of the growth in manufacturing. Toward this
end, we utilize a standard growth accounting framework in which output is defined as
manufacturing shipments adjusted for price changes—termed real gross output. We decompose
growth into the parts resulting from the growth of inputs to production—capital, labor, services,
energy, and materials inputs—and MFP growth, which is computed as a residual. In other words,
MFP growth is the part of output growth that cannot be accounted for by the growth of factor
inputs, and therefore represents the returns to all factors of production.151
As mentioned, a novel feature of our analysis is that we distinguish between domestically
sourced and foreign materials inputs. Specifically, we use unpublished BEA data on the value of
149
With the BEA’s May 2010 comprehensive revision to the annual industry accounts, manufacturing output now
expands at a slightly faster rate during this period. The analysis throughout this paper is based upon the previous
vintage of these data published in 2009.
150
This perspective is illustrated by Executive Office of the President (2009), which emphasizes the strength of
output and productivity and growth of U.S. manufacturers vis-à-vis the aggregate economy and manufacturers in
other industrialized countries and which largely attributes employment declines to productivity growth. Recent
articles in the popular press also have advanced this view (e.g., Economist 2005; Murray 2009).
151
See Jorgenson, Gollop, and Fraumeni (1987) and Hulten (2009) for more on the growth accounting methodology,
its early development, and current applications. The industry-level data for output, materials, energy, and services
come from the BEA’s GDP-by-industry accounts. Capital services inputs for are derived from the BEA’s fixed asset
accounts. The labor input is based on industry-level hours worked from the national income and product accounts,
adjusted for changes in the worker composition effects using wage data from the Census Bureau’s county business
patterns.
292
imports and imported input prices at a detailed commodity level to distinguish between the
growth of domestic and imported materials inputs.
Using these data, we compute MFP growth for manufacturing as the growth in real gross output
∧
∧
∧
∧
( Q ) less a weighted average of the growth rate of labor ( L ), capital ( K ), energy ( E ), services
∧
∧
∧
( S ), domestic materials ( M D ), and foreign materials ( M F ) inputs:
(1)
∧
∧
∧ ⎡
∧
∧
∧
∧
⎤
L
K
E
S
mD
D + wm F
F⎥
w
w
w
w
w
=
−
+
+
+
+
⎢
Q
L
K
E
S
M
M
MFP
⎢⎣
⎥⎦
∧
The weights (w) on each input represent the input’s share of total input costs. Any error in the
measurement of input growth—including errors that result from biased price indexes used to
deflate the inputs—will directly result in an error in the measurement of productivity growth.
Equation (1) can be rearranged to obtain an identity in which output growth is decomposed into
MFP growth and contributions from the growth of factor inputs. Table 1 provides the results of
this decomposition for manufacturing and selected industry breakouts from 1997 to 2007. Note
that the figures in column (1), which represent the average annual real output growth rate over
the period, equal the sum of the figures in columns (2) through (8), which represent the
contributions to output growth from MFP and from the growth of the indicated inputs.
Several striking findings emerge from this table. One is the strong MFP growth. The contribution
to real output growth from MFP actually exceeds real gross output growth, indicating that MFP
can account for all of the growth in real gross output over the decade. Capital, purchased
services, and materials all play important, albeit more modest, roles, while the contribution of
labor is negative and large, reflecting the steep employment declines during the period.152
Columns (7) and (8) in Table 1 provide a clear picture of the rapid pace of structural change
currently under way in U.S. manufacturing. During the period, the contribution of domestically
supplied materials inputs fell, while that of imported materials inputs greatly expanded,
reflecting the substitution of foreign for domestic intermediate inputs. The growth of imported
intermediate inputs, to some degree, will also reflect the direct substitution of imported goods for
domestic labor and capital. To see this, consider the case in which a firm previously produced an
intermediate input and final product internally, but now sources that input from a foreign
supplier. In this instance, gross output will not change, but imported materials inputs will rise
and the labor and capital previously used to produce the input will fall.
For all manufacturing, the contribution of imported materials inputs to output growth was greater
than that of any other factor of production and was more than double the contribution from
capital. For manufacturing excluding the computer industry, imported materials account for 60
percent of the growth during this period.
152
The growth accounting results in Table 1 reflect the authors’ calculations and rely on a different methodology
than what is used by the BLS. However, these two salient features of the data are also observed in the BLS
estimates. A full reconciliation of the two approaches appears in Houseman et al. (2010).
293
Another striking result in Table 1 is that computer and electronic products manufacturing—which
includes computers, semiconductors, and telecommunications equipment—accounts for most of the
output and productivity growth in manufacturing over the period.153 Output and productivity growth in the
computer industry averaged 7.4 and 6.8 percent per year, respectively, compared to growth of only about
0.5 percent for output and 0.7 percent for MFP in the rest of manufacturing. The extraordinary
productivity and output growth in computers reflects, to a large degree, technological improvements of
the products produced and output price deflators that, when properly adjusted for product improvements,
are often falling rapidly.154
Throughout the decade, the computer industry’s share of manufacturing value added remained
relatively constant at around 10 percent. Because manufacturing output and productivity
statistics are strongly affected by the computer industry, which represents such a small share of
the sector, researchers should be cautious in drawing general inferences about manufacturing
from the aggregate numbers. It also bears note that statistics on industry output and productivity
growth should be interpreted in relation to growth in demand for the industry’s products. Indeed,
in spite of rapid value added and productivity growth of computers and electronic products
manufacturing during the decade, the trade deficit within this product group greatly widened and
substantial offshoring of components of the industry occurred (Brown and Linden 2005; Linden,
Dedrick, and Kraemer 2009).
Because statistics on labor productivity, defined as output per hour worked, are widely used in
research and policy analyses, it is also of interest to consider the relationship between labor
productivity growth and offshoring. In the official BLS labor productivity release, manufacturing
output includes imported intermediates but excludes intermediates sourced from within the
domestic manufacturing sector. As a result, shifts in sourcing from a domestic to a foreign
supplier do not offset each other, mechanically increasing labor productivity.155 To this point,
Eldridge and Harper (2009) find that imported intermediate materials explain 20 percent of the
growth in manufacturing labor productivity from 1997 to 2006. We find that the contribution to
manufacturing labor productivity from imported materials inputs significantly accelerated over
the period.
Although Table 1 documents the substantial growth in offshoring during the period, it nevertheless likely
understates the true magnitude of the phenomenon. Our focus below concerns the systematic upward bias
in the price indexes used to deflate intermediate materials. We could not account for the measurement of
two additional factors, which likely also impart an upward bias: 1) imported capital inputs, such as
computers and machinery, have exhibited substantial gains in import penetration; and 2) imported
services inputs (i.e., services offshoring) have accelerated in recent years, albeit from a very low level.156
153
Similar findings have been reported in other studies. See, for example, Jorgenson, Ho, and Stiroh (2008) and
Oliner and Sichel (2000). See also Oliner, Sichel, and Stiroh (2007) and Syverson (2010) for more in-depth reviews
of recent research on U.S. productivity.
154
The BLS uses hedonic methods to adjust prices in the computer industry. For a review of these, see Wasshausen
and Moulton (2006).
155
This could also occur if a firm imports an intermediate input it previously produced internally. In this case, output
will not change, but the labor input used to produce that intermediate input will fall.
156
See Cavallo and Landry (2010) for a discussion of imported capital goods, and Eldridge and Harper (2009) and
Yuskavage, Strassner, and Maderios (2008) and for estimates of services offshoring.
294
BIAS TO PRICE INDEXES FROM OFFSHORING
Understanding why offshoring likely results in biases to the price indexes used to deflate inputs
requires some background on the relevant price programs. In addition to the Consumer Price
Index (CPI), the BLS constructs separate price indexes for imports, exports, and domestically
produced goods. Just as the BLS constructs the CPI to measure the rate of price change of goods
and services purchased by consumers, the BEA constructs input price indexes to measure the rate
of price change of inputs to production purchased by businesses. The BEA constructs industryspecific input price indexes using import and domestic price indexes in conjunction with
information on each industry’s input structure from the input-output tables. The import price data
come from the BLS’s International Price Program (IPP), which surveys importing establishments
on the prices paid for imports of a detailed product. For domestic materials prices, the BEA
primarily uses the Producer Prices Index (PPI) in which BLS surveys domestic producers on the
prices they receive for a sample of products.157
The BLS takes great care to ensure that it is pricing the same item over time, and thus that price
indexes are based on “apples-to-apples” comparisons. Each observation used in the construction
of a particular price index represents the period-to-period price change of an item as defined by
very specific attributes and reported by a specific establishment. These methods mean, however,
that price indexes neither capture the price changes associated with the entry of a low-cost
supplier nor the level differences in prices which drive its subsequent market share expansion.
As mentioned, this problem has been studied extensively in the CPI literature, where it has been
dubbed “outlet substitution bias” (Diewert 1998, Hausman 2003, and Reinsdorf 1993).
Figure 4 presents a stylized depiction of the problem in the context of offshoring. The BLS
measures the price change from period t to t+1 of a specific imported product at a specific
importer in the IPP, and it measures the price change from period t to t+1 of a specific product
produced by a specific domestic producer in the PPI. Neither the IPP nor the PPI captures the
price drop (d) that occurs when businesses shift from a high-cost domestic to a low-cost foreign
supplier. The input price index is, in essence, a weighted average of period-to-period changes
measured in the IPP and the PPI, and thus the price drop from offshoring is missed.158
To further illustrate how offshoring can impart a bias into the input price index, Table 2 provides
a hypothetical numerical example. Suppose that pharmaceutical companies purchase a common
chemical compound, “obtanium,” from a domestic supplier at $10 per ounce. A Chinese supplier
enters the market and sells obtanium for $6 per ounce. As the new, lower-cost source becomes
known, its reliability is established, and contracts with the domestic supplier expire, U.S.
pharmaceutical companies begin shifting their purchases to the Chinese supplier. For simplicity,
we assume that the domestic and foreign dollar prices of obtanium remain the same throughout
the period.159 Even if the BLS picks up the Chinese import of obtanium in its import prices
157
For more information on the BLS price index computations, see Chapters 14 and 15 in BLS (2009).
Although our empirical focus is on the substitution of imported for domestic inputs, a bias would also occur with
the entry and market share expansion of a new low-cost domestic supplier.
159
Because prices are often contractually set for periods of time, this simplifying assumption of price stickiness is
not unrealistic. Nakamura and Steinsson (2009) document that 40 percent of prices on imported items never change
for the entire duration they are in the BLS sample.
158
295
sample without a lag, it will not capture the input price drop enjoyed by drug manufacturers at
the time of the switch.
The input price index, as computed by the statistical agencies, is a weighted average of the
domestic and import index, and, in our example, does not change. The correct index, however,
would capture the period-to-period change of the average price that U.S. companies pay for
obtanium and falls by 12 percent. More rapid introduction of new suppliers into the BLS
sampling frame or more frequent sampling of prices—common suggestions for improving price
statistics—will not address this particular problem.
The bias to the price index arises because neither the U.S. producer nor the U.S. importer can
report the price drop that buyers experience when they shift their purchases from domestic to
foreign suppliers. To address this problem, the BLS has proposed that an input index be
constructed based on a survey of purchasers (Alterman 2009).160 In theory, the buyers could
accurately report the period-to-period changes in the price they pay for specific inputs,
irrespective of source.
Diewert and Nakamura (2010) characterize the bias to the input price index resulting from a shift
in input suppliers. The upward bias (B) to the rate of inflation in the input price index (1+i) is
proportional to the share captured by the low-cost supplier over the period (s) and the percentage
difference in the prices of the low- versus the high-cost supplier—or discount—(d):
(2)
B = (1 + i ) sd
Returning to our obtanium example, over the period there is no measured inflation (i.e., i equals
zero), the low-cost supplier captures a 30 percent market share, and the discount from the lowcost supplier is 40 percent. Whereas the measured rate of price change is zero, the true rate of
price change for that input is -0.12, or negative 12 percent. The characterization of the bias to the
input price index in Equation (2) is identical to the characterization of the bias to the CPI
(Diewert 1998). It is the same problem manifested in a different index. 161
As shown in Figure 5, the problem is evident in the input price indexes used by the BEA to
deflate intermediate materials inputs. If price indexes were accurately capturing the cost savings
to businesses that presumably underlie the recent share growth of imported intermediates, then
the growth of the import price index should be slower than the domestic price index, indicating a
fall in the price of imported relative to domestic inputs. Instead, the foreign price deflator for
intermediate materials rose faster than the domestic deflator. The differential between foreign
and domestic materials price deflators is especially apparent beginning in 2002, coincident with
the rapid rise of imports from China.
160
The proposed input cost index is still at the concept and design stage at the BLS.
Although the goods in this model are treated as homogeneous, Diewert (1998) provides a simple extension to
where the goods are different qualities. In this case, the discount represents the price differential adjusted for quality.
161
296
EVIDENCE OF COST SAVINGS FROM OFFSHORING
No comprehensive evidence exists on the magnitude of the cost savings from shifts in
sourcing—i.e., the discount, d, in Equation (2). A few case studies, however, provide some
evidence for selected products and industries. Byrne, Kovak, and Michaels (2010) find sizable
cross-country differences in the prices of semiconductor wafers with identical specifications. On
average, they find that, compared to prices of semiconductor wafers produced in U.S. foundries,
prices were on average about 40 percent lower in China and about 25 percent lower in
Singapore. Klier and Rubenstein (2009) find that offshoring aluminum wheel production to
Mexico lowered overall costs by 19 percent and processing costs by 36 percent.
The different samples in the IPP and PPI do not permit a direct comparison of prices for
domestic and imported items. However, such a comparison is possible among imported products
originating in different countries. Products from intermediate and, especially, developing
countries were gaining market share not only vis-à-vis the United States but also other advanced
countries. On the grounds that production cost structures are likely to be more similar between
the United States and other advanced countries, systematic import price differentials between
products from advanced versus developing and intermediate countries may be informative about
the size of the discount relative to U.S. goods.
Figure 6 shows the average percentage differences between imported products from advanced
and developing countries and between advanced and intermediate countries as recorded in the
BLS microdata underlying the IPP. The position of each bubble represents the size of the
discount for a single North American Industry Classification System (NAICS) 4-digit category in
manufacturing. The size of each bubble is proportional to the gain in U.S. market share for
developing and intermediate countries within each category. In almost all cases, the discounts are
negative, indicating lower prices in developing and intermediate countries compared to advanced
countries. In many cases, these discounts appear to be quite sizable. Further, the size of a
discount is negatively correlated with a gain in market share, indicating that the larger the
developing or intermediate country’s price differential, the greater the U.S. market share
captured. Products on the left side of the figure (i.e., food, beverages, textiles, and apparel) are
characterized by smaller discounts and share gains, whereas products to the right (i.e.,
machinery, electronics, semiconductors, and transportation) are characterized by larger discounts
and share gains.
An important caveat to this figure is that even within very detailed product codes there may be
considerable heterogeneity that may explain at least some of the price differentials. Returning again to our
example in the previous section, if obtanium is a differentiated product and the Chinese version is of a
lower quality than that from Japan, then Chinese obtanium should trade at a discount relative to a variety
that is not strictly comparable. We adopt various methods to control for possible heterogeneity. In
particular, one method restricts the import price sample to cases in which there is a newly observed price
for an incumbent importing firm within the same detailed product code. In this way, we are able to narrow
the sample to instances in which an importer appears to be switching sources of a specific product from a
297
supplier in an advanced country to one in a developing or intermediate country. The observed price
differentials are somewhat smaller but still sizable when we limit our sample in this way.162
The evidence from case studies and from comparisons of import prices is consistent with reports
of large discounts in the business literature. For example, in 2004 Business Week reported that
prices of imported goods from China typically were 30 to 50 percent lower than the prices for
comparable products produced in the United States, and that the discounts were sometimes
higher (Engardio and Roberts 2004). Similarly, a McKinsey & Company (2006) study cited cost
savings from production of electronic equipment in China of between 20 and 60 percent.
Estimates of the savings from offshoring auto parts production to Mexico are generally in the
range of 20 to 30 percent (see discussion in Kinsman 2004). In sum, although no systematic data
exist, a variety of evidence points to large cost savings from offshoring.
The above-mentioned price differentials could be the result of numerous factors, such as labor
costs, industrial policy, or disequilibrium in exchange rate markets. For instance, the
Manufacturers Alliance of the National Association of Manufactures provides estimates of
manufacturing labor costs, adjusted for productivity, for major U.S. trading partners as compared
to the United States. Their estimates of large labor cost savings—58–72 percent lower in China
and 22–62 percent in Mexico from 2002 to 2009—are consistent with the large product discounts
reported in research and in the business press (Leonard 2008). Also consistent with the evidence
of cost savings from offshoring are estimates that the Chinese renminbi may be significantly
undervalued relative to the dollar, perhaps by as much as 40 percent (Bergsten 2010, Cline and
Williamson 2010).
BIAS TO PRODUCTIVITY AND VALUE ADDED FROM OFFSHORING
We implement the bias correction to input prices in Equation (2) and simulate the effects of the
bias on MFP and value-added growth. Figure 7 illustrates what the bias to industry-level
materials deflators would be if the true import discounts match those derived from the full
sample of import price microdata. The vertical distance between each point and the 45 degree
line represents the size of the bias. For all manufacturing, if the true import discount can be
approximated by the IPP microdata, then the cumulative price growth of 20 percent between
1997 and 2007 would overstate the bias-corrected inflation rate by a full 9 percentage points.
Thus, once we account for offshoring bias, the materials costs faced by U.S. manufacturers
would only have risen at half the rate reported in official estimates. This, in turn, would imply
that the real use of materials by U.S. manufacturers was twice as large as reported. With more
production being generated by purchased materials, both productivity and value added would be
diminished. Also shown in Figure 7 are the implied industry-level cumulative price changes
under this set of assumptions about import price discounts.
162
Houseman et al. (2010) describe this approach in greater detail. We also attempt to account for unobserved
differences in product characteristics using an econometric model informed by estimates of product-level quality
from Mandel (2010). We find that the price dispersion across source countries decreases but remains substantial. In
short, it is unlikely that product differentiation accounts for the large, persistent price differences across countries.
298
By how much might the productivity statistics be overstated from failing to account for
offshoring? The top panel of Table 3 presents alternative estimates of MFP growth based upon
different assumptions about the import discount. Column (1) restates our baseline MFP results
from Table 1, while column (2) presents estimates in which all commodities—both domestic and
imported—have been deflated with domestic deflators provided to us by the BEA. Since our
alternative materials deflators are derived by adjusting domestic commodity prices (primarily
PPIs), the estimates in column (2) should be viewed as the appropriate reference or “jumpingoff” point for gauging the extent of offshoring bias. In other words, they show what MFP would
be if the rate of price inflation for imported commodities was the same as for their domestic
counterparts. This assumption in Equation (2) is maintained in order to hone in on the impact of
the level difference in prices between imported and domestic commodities.
For the entire manufacturing sector, deflating imported materials with domestic prices reduces
MFP growth by a bit less than 0.1 percentage point, from 1.30 percent in our baseline scenario to
1.23 percent. Almost all of this change owes to differences in the price deflators used for
imported and domestic semiconductors. In other words, prices for imported semiconductors—a
product used heavily by the computer and electronic products industry—fell less rapidly than
their domestic counterparts. The discrepancies are especially evident in the early years of our
data and appear to be the result of inconsistent adjustment of imported and domestic
semiconductor prices for quality improvements. Although not the focus of our paper, the drop in
MFP between columns (1) and (2) likely represents an additional modest bias.163
Columns (3) and (4) present MFP estimates that have been adjusted for offshoring using our
micro evidence on the import discount. We report estimates using product-level discounts based
on the entire microdata sample (“full sample”) and on a sample limited to instances where
importers appear to shift from suppliers in advanced counties to ones in developing or
intermediate countries (“switchers”). In columns (5) and (6), we estimate MFP using import
discounts informed by the business press and available case study evidence, applying these
discounts uniformly across commodities. The column labeled “50/30” assumes discounts of 50
percent for developing countries and 30 percent for intermediate countries, whereas the column
labeled “30/15” assumes discounts of 30 percent for developing countries and 15 percent for
intermediate countries. These represent discounts on the high and low end, respectively, of those
found in the case study and business literature.
On balance, for the entire manufacturing sector, we find that adjusting for offshoring lowers
MFP growth by an additional 0.1–0.2 percentage point, which implies average annual
productivity growth is reduced between 6 and 14 percent. These numbers are fairly significant,
as a 0.1 percent average annual growth rate for MFP roughly equals the average annual
contribution of the capital stock to manufacturing growth during this period.
If we exclude the contribution of the computer and electronic products industry, correcting for
offshoring results in larger percentage adjustments to MFP, which falls from 0.67 percent in
163
Because of the high import penetration in semiconductors and other high-tech products, consistently adjusting
domestic and import prices for product improvements is important for the accuracy of industry and national income
statistics, though difficult owing to lack of product detail, particularly for imports. Addressing this problem has
resulted in substantial revisions to the national accounts statistics in the past (Grimm 1998).
299
column (2) to between 0.52 percent (column 3) and 0.58 percent (column 4); in other words, the
reduction in MFP widens to as much as 22 percent. The results for the case study scenarios
shown in columns (5) and (6) are quite consistent with our results based on IPP microdata.
What about the likely range of bias to value added? Recall that value added nets out intermediate
inputs from an industry’s shipments, and therefore represents the additional product produced in
an industry.164 If, as suggested by Figure 7, the actual amount of intermediate materials used by
U.S. manufacturers has been larger than what is contained in the official statistics, then value
added has been overstated as well. This implies that a larger share of the sector’s production has
simply been final assembly, and relatively less of domestic manufacturing shipments are
contributing value to the overall economy.
The BEA derives indexes for industry-level value added using the double-deflation method in
which real value added is computed as the difference between real gross output and real
intermediate inputs. We replicate this double-deflation procedure using our adjusted measures of
real purchased materials. We therefore derive the implied value of real value added –associated
with published measures of gross output, energy, and services and our adjusted measures of
purchased materials inputs.
The bottom panel of Table 3 presents alternative estimates for value added based on our different
assumptions on the import discount. Of note, the unadjusted average growth rate in value added
for all manufacturing is about 3 percent, while the annual growth rate for manufacturing,
excluding the computer sector, is less than one-third of this size—about 0.9 percent. The annual
growth rate for the computer industry exceeds 20 percent. As shown in columns (3) through (6),
our simulations indicate that value-added growth for all manufacturing was overstated by 0.2–0.5
percentage point, or about 7–18 percent of the growth. Excluding computers, real value-added
growth for manufacturing is biased by 0.2–0.4 percentage point, which accounts for 21–49
percent of the growth. 165 The annual growth rate of real value added for manufacturing
excluding computers falls under a half percent per year in some of our adjusted estimates, while
that for nondurable goods turns negative for all of our adjusted estimates.
IMPLICATIONS FOR DATA AND RESEARCH
The above analysis focuses on biases to manufacturing productivity and value added from the
substitution of lower-cost imported for domestic materials inputs. Such biases, however, may
also arise from the offshoring of other inputs and affect statistics for other sectors and for the
aggregate economy.166 In the 2000s, sizable import penetration by developing countries occurred
in computers and machinery products, which largely are treated as capital inputs in the industry
accounts. Price drops accompanying the substitution of imported for domestic capital equipment
164
Gross domestic product can therefore be derived as the sum of value added across all sectors of the economy.
In addition to the “switchers” estimates, we attempted to adjust for unobserved differences within detailed
product codes using econometric techniques. These estimates do not alter the qualitative results of our analysis, and
imply bias adjustments to MFP and value added roughly in line with the “30/15” estimates in Table 3.
166
Reinsdorf and Yuskavage (2009) examine pricing in selected consumer goods and provide preliminary evidence
of biases to GDP from import growth. Biases to price indexes from offshoring and their implied biases to GDP
growth also have been covered in the business press (see Mandel 2007, 2009).
165
300
would not be captured in capital price deflators, possibly leading to an understatement of the
growth of capital services and an overstatement of growth in MFP and real value added.
The same problem arises from services offshoring. Collecting accurate price information on
services trade is complicated by the fact that the level of detail in services sector data is quite
limited (Jensen 2009, Norwood et al. 2006,Sturgeon et al. 2006, ), and that the BLS international
prices program does not cover business services imports and exports. If services offshoring were
to expand rapidly in the near future, as some predict, the absence of accurate price deflators
could impart significant biases in official statistics.
More generally, the Shumpetarian dynamic by which low-cost producers enter and capture
market share from incumbents is an important mechanism by which prices change, but is a
dynamic largely missed in price indexes. Although we have focused on the substitution of lowcost foreign for domestic inputs because of the recent empirical importance of offshoring, the
entrance and market share expansion of low-cost domestic suppliers is an important aspect of
firm dynamics in the United States and also would impart biases to price indexes.167 As
mentioned above, a proposal to construct an input price index based on a survey of purchasers, if
implemented by BLS, would address the biases to the industry statistics from all shifts in
sourcing (Alterman 2009).168
The growth of low-cost imports has spurred numerous studies to assess their effects on the U.S.
economy and its workers. Biases to price indexes that arise from offshoring affect the accuracy
not only of the industry statistics, but also of analyses based on them. Because such import
growth will be understated in real terms and, to some degree, will be manifested as false
productivity gains, studies may underestimate the true effects of import growth.
The pace of globalization is unlikely to abate in the near future, underscoring the need for
reliable economic statistics to understand its effects and formulate policy responses. The biases
to price indexes discussed in this paper are emblematic of a broader set of measurement
problems that arise from the growth of globalization (Feenstra and Lipsey 2010, Houseman and
Ryder 2010). Understanding the effects of globalization requires better data, including, quite
critically, better price deflators.
167
See Foster, Haltiwager, and Syverson (2008) for evidence that entrants, on average, have higher physical
productivity and offer lower prices than incumbent firms.
168
The proposed index, which would not distinguish source country, would capture price changes from shifts in
sourcing among domestic suppliers, among domestic and international suppliers, and among international suppliers.
Although the input price index would address biases in the industry statistics, it would not address biases in
published statistics on GDP growth, which are based on expenditure, not industry value-added, data.
301
References
Alterman, William. 2009. “Producing an Input Price Index.” Paper prepared for the conference
DC, November, 6–7
Bergsten, C. Fred. 2010. “Beijing Is Key to Creating More U.S. Jobs.” Foreignpolicy.com.
http://www.foreignpolicy.com/articles/2010/04/14/china_the_job_killer?print=yes&hidec
omments=yes&page=full (accessed June 10, 2010).
Brown, Clair, and Greg Linden. 2005. “Offshoring in the Semiconductor Industry: A Historical
Perspective.” In Brookings Trade Forum: 2005: Offshoring White-Collar Work, Lael
Brainard and Susan M. Collins, eds. Washington, DC: Brookings Institution Press, pp.
279–333.
Bureau of Labor Statistics. 1997. “International Price Indexes.” BLS Handbook of Methods, 154–
159. Washington, DC.: U.S. Government Printing Office.
Bureau of Labor Statistics. 2008. “Producer Prices.” BLS Handbook of Methods, 1–12.
Washington, DC.: U.S. Government Printing Office.
Bureau of Labor Statistics. 2009. “International Comparisons of Manufacturing Productivity and
Unit Labor Cost Trends, 2008.” News Release.
http://www.bls.gov/news.release/pdf/prod4.pdf
Byrne, David, Brian Kovak, and Ryan Michaels. 2010. “Offshoring and Price Measurement in
the Semiconductor Industry.” Paper prepared for the conference “Measurement Issues
Arising from the Growth of Globalization,” held in Washington, DC, November, 6–7,
2009.
Cavallo, Michele, and Anthony Landry. 2010. “The Quantitative Role of Capital Goods Imports
in U.S. Growth.” American Economic Review 100(2): 78–82.
Cline, William R., and John Williamson. 2010. “Notes on Equilibrium Exchange Rates: January
2010.” Peterson Institute for International Economics Number PB10-2. Washington, DC:
Peterson Institute for International Economics.
Corrado, Carol, and Joe Mattey. 1997. “Capacity Utilization.” Journal of Economic Perspectives
11(1): 151–167.
Diewert, W. Erwin. 1998. “Index Number Issues in the Consumer Price Index.” Journal of
Economic Perspectives 12(1): 47–58.
Diewert, W. Erwin, and Alice Nakamura. 2010. “Bias in the Import Price Index Due to
Outsourcing: Can It be Measured?” Paper prepared for the conference “Measurement
303
Issues Arising from the Growth of Globalization,” held in Washington, DC, November,
6–7, 2009.
Economist. 2005. “Finance and Economics—Manufacturing Employment—Industrial
Metamorphosis.” 377(8446): 69.
Eldridge, Lucy P., and Michael J. Harper. 2009. “Effects of Imported Intermediate Inputs on
Productivity.” Paper prepared for the conference “Measurement Issues Arising from the
Growth of Globalization,” held in Washington, DC, November, 6–7.
Engardio, Pete, and Dexter Roberts. 2004. “The China Price.” Business Week, December 6.
http://www.businessweek.com/magazine/content/04_49/b3911401.htm (accessed June 9,
2010).
Executive Office of the President of the United States. 2009. “A Framework for Revitalizing
American Manufacturing.” Washington, DC: U.S. Government Printing Office.
Feenstra, Robert, and Robert Lipsey. 2010. Report on the State of Available Data for the Study of
International Trade and Investment. Report prepared for the Committee on Economics
Statistics of the American Economic Association.
Foster, Lucia, John Haltiwanger, and Chad Syverson. 2008. “Reallocation, Firm Turnover and
Efficiency: Selection on Productivity or Profitability?” American Economic Review
98(1): 394–425.
Grimm, Bruce T. 1998. “Price Indexes for Selected Semiconductors.” Survey of Current
Business 78(2): 8–24.
Hausman, Jerry. 2003. “Sources of Bias and Solutions to Bias in the Consumer Price Index.”
Journal of Economic Perspectives 17(1): 23–44.
Helper, Susan. 2008. “Renewing U.S. Manufacturing: Promoting a High-Road Strategy.” EPI
Briefing Paper 212. Washington, DC: Economic Policy Institute.
Houseman, Susan, Christopher Kurz, Paul Lengermann, and Benjamin Mandel. 2010.
“Offshoring Bias: The Effect of Import Price Mismeasurement on Manufacturing
Productivity and Value Added Growth.” Mimeo.
Houseman, Susan, and Kenneth Ryder. 2010. Measurement Issues Arising from the Growth of
Globalization. Report prepared for the U.S. Bureau of Economic Affairs, Washington,
DC
Hulten, Charles R. 2009. “Growth Accounting.” NBER Working Paper No. 15341. Cambridge,
MA: National Bureau of Economic Research.
Jensen, J. Bradford. 2009. “Measuring the Impact of Trade in Services: Prospects and
Challenges.” Paper prepared for the conference “Measurement Issues Arising from the
Growth of Globalization,” held in Washington, DC, November, 6–7.
304
Jorgenson, Dale W., Frank M. Gollop, and Barbara M. Fraumeni. 1987. Productivity and U.S.
Economic Growth. Cambridge, MA: Harvard University Press.
Jorgenson, Dale W., Mun S. Ho, and Kevin J. Stiroh. 2008. “A Retrospective Look at the U.S.
Productivity Growth Resurgence.” Journal of Economic Perspectives 22(1): 3–24.
Kinsman, Michael. 2004. “Exporting Work: Outsourcing That Once Sent Low-Skill Jobs to
Mexico Is Now Sending Some of San Diego’s High-Skilled, High-Wage Jobs to India,
China and Elsewhere.” San Diego Tribune, April 4.
http://64.9.213.241/files/old/news/articles/san_diego_tribune_4_4_04.pdf (accessed June
9, 2010).
Klier, Thomas H., and James M. Rubenstein. 2009. “Imports of Intermediate Parts in the Auto
Industry—A Case Study.” Paper prepared for the conference “Measurement Issues
Arising from the Growth of Globalization,” held in Washington, DC, November 6–7.
Leonard, Jeremy A. 2008. “The Tide is Turning: An Update on Structural Cost Pressures Facing
U.S. Manufacturers.” Published jointly by the Manufacturers Alliance/MAPI and The
Manufacturing Institute. http://institute.nam.org/view/2001005065073451188/info
(accessed June 10, 2010).
Linden, Greg, Jason Dedrick, and Kenneth L. Kraemer. 2009. “What the iPod Tells Us.”
http://delong.typepad.com/egregious_moderation/2009/12/greg-linden-jason-dedrickand-kenneth-l-kraemer-what-the-ipod-tells-us.html (accessed June 10, 2010).
Mandel, Michael. 2007. “The Real Costs of Offshoring.” Business Week, June 18.
http://www.businessweek.com/magazine/content/07_25/b4039001.htm (accessed June 9,
2010).
———. 2009. “Growth: Why the Stats Are Misleading.” Business Week, June 3.
http://www.businessweek.com/magazine/content/09_24/b4135000594984.htm (accessed
June 9, 2010).
Mandel, Benjamin R. 2010. “Heterogeneous Firms and Import Quality: Evidence from
Transaction-Level Prices.” Board of Governors of the Federal Reserve System,
International Finance Discussion Paper No. 991. Washington, DC: Board of Governors of
the Federal Reserve System.
McKinsey & Company. 2006. “China and India as Global Supply Points: A Business
Opportunity!” Consensor Results. Presentation Pudong Shangri-La Hotel, Shanghai, May
22–24.
Murray, Sara. 2009 “U.S. Manufacturing Productivity Jumps.” Wall Street Journal, October 23, A:2.
Nakamura, Emi, and Jón Steinsson. 2009. “Lost in Transit: Product Replacement Bias and
Pricing to Market.” New York: Columbia University
http://www.columbia.edu/~en2198/papers/ippsubs.pdf (accessed June 10, 2010).
305
New America Foundation. 2010. “America Needs a Manufacturing Strategy.” (February 3).
http://www.newamerica.net/publications/policy/america_needs_a_manufacturing_
strategy.
Norwood, Janet L., Carol Carson, Manuel Deese, Norman J. Johnson, Franklin S. Reeder, and
John E. Rolph. 2006. Offshoring: How Big Is It? Report of the Panel of the National
Academy of Public Administration for the U.S. Congress and the Bureau of Economic
Analysis, National Academy of Public Administration, Washington, DC.
Oliner, Stephen D., and Daniel E. Sichel. 2000. “The Resurgence of Growth in the Late 1990s: Is
Information Technology the Story?” Journal of Economic Perspectives 14(4): 3–22.
Oliner, Stephen D., Daniel E. Sichel, and Kevin J. Stiroh. 2007. “Explaining a Productive
Decade.” Brookings Papers on Economic Activity 1: 81–137.
Pisano, Gary P., and Willy C. Shih. 2009. “Restoring American Competitiveness.” Harvard
Business Review July-August: 114–125.
Pollin, Robert, and Dean Baker. 2010. “Reindustrializing America: A Proposal for Reviving U.S.
Manufacturing and Creating Millions of Good Jobs.” New Labor Forum 19(2): 17–34.
Reich, Robert B. 2009. “Made in America: Manufacturing Jobs are Never Coming Back.”
Forbes.com: http://www.forbes.com/2009/05/28/robert-reich-manufacturing-businesseconomy.html (accessed June 10, 2010)
Reinsdorf, Marshall. 1993. “The Effect of Outlet Price Differentials in the U.S. Consumer Price
Index.” In Price Measurements and Their Use, Murray F. Foss, Marylin E. Manser, and
Allan H. Young, eds. Chicago: University of Chicago Press, pp. 227–254.
Reinsdorf, Marshall, and Robert Yuskavage. 2009. “Are There Unmeasured Declines in Prices of
Imported Final Consumption Goods?” “Measurement Issues Arising from the Growth of
Globalization,” held in Washington, DC, November, 6–7.
Sturgeon, Timothy J., Frank Levy, Clair Brown, J. Bradford Jensen, and David Weil. 2006.
“Why We Can’t Measure the Economic Effects of Services Offshoring: The Data Gaps
and How to Fill Them.” Services Offshoring Working Group Final Report. Cambridge,
MA: Massachusetts Institute of Technology, Industrial Performance Center.
http://web.mit.edu/ipc/publications/pdf/IPC_Offshoring_Report.pdf (accessed June 10,
2010).
Surdna Foundation. 2010. “Revitalizing America’s Manufacturing Sector.” Surdna.org.
http://www.surdna.org/what-we-fund/sustainable-environments/sustainableenvironments-whats-new/203-revitalizing-americas-manufacturing-sector-summary-ofsurdna-roundtable.html.
Syverson, Chad. 2010. “What Determines Productivity?” NBER Working Paper No. 15712.
306
Wasshausen, David B., and Brent R. Moulton. 2006. “The Role of Hedonic Methods in
Measuring Real GDP in the United States. Prepared for the 31st Committee on Statistical
Information in the Economic and Social Sphere (CEIES) seminar “Are We Measuring
Productivity Correctly?” held in Rome, October 12–13.
Imported Services in BEA’s Industry Accounts.” In International Flows of Invisibles:
Trade in Services and Intangibles in the Era of Globalization, Marshall Reinsdorf, and
Matthew Slaughter, eds. Chicago: University of Chicago Press, pp. 247–288.
307
Figure 1: The Import Share of Material Inputs Used by U.S. Manufactures
0.3
0.25
0.25
0.2
Import Share
0.2
0.15
Developing
0.15
Intermediate
0.1
0.1
0.05
Advanced
0
1997
1998
0.05
1999
2000
2001
2002
2003
2004
2005
2006
NOTE: Imported intermediates are decomposed into their source country of origin and plotted as
their portion of the share of imported intermediate in total materials use by the manufacturing
sector.
Source: BEA Annual Industry Accounts and Import Microdata
Figure 2: Manufacturing Employment, 1960-2009
20
19
Employment (millions)
18
17
16
15
14
13
12
11
Source: BLS
308
20
08
20
05
20
02
19
99
19
96
19
93
19
90
19
87
19
84
19
81
19
78
19
75
19
72
19
69
19
66
19
63
19
60
10
0
2007
Figure 3: Manufacturing: Labor Productivity and Employment Relative to Total
35
1.4
30
Employment Share
(percent)
1.3
Productivity Ratio (right
axis)
1.2
25
1.1
1
20
0.9
15
0.8
0.7
10
0.6
20
08
20
05
20
02
19
99
19
96
19
93
19
90
19
87
19
84
19
81
19
78
19
75
19
72
19
69
19
66
0.5
19
63
19
60
5
NOTE: Productivity series is calculated as ratio of manufacturing output per worker as a fraction of total
nonfarm business labor productivity. Employment share is ratio of manufacturing employment to total
employment.
SOURCE: BLS
309
Figure 5: Baseline Input Price Indexes for the Manufacturing Sector
1.5
Imported Materials
Total Materials
1.4
Domestic Materials
1.3
1.2
1.1
1
0.9
0.8
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
NOTE: The total materials deflator is from the BEA’s Annual Industry Accounts, while the imported
materials deflator is an aggregate of confidential BEA commodity price data. The implied domestic materials
price deflator is computed from the total and imported materials price deflators.
Figure 6
NOTE: The import price discount for each NAICS category averages across many underlying detailed
product groups classified according to the U.S. Harmonized System (HS 10-digit) for the months
September 1993–May 2007. Within an HS group, the developing (or advanced) country discount is the
average of individual item prices exported from developing (or advanced) countries relative to a
geometric mean of advanced country transaction prices.
SOURCE: BLS
310
Figure 7: Materials Cost Inflation for U.S. Manufacturing Industries
NOTE: The figure contrasts the materials cost inflation as published by the BEA with an adjusted measure derived
from IPP micro data and Census foreign trade shares. For each manufacturing industry, and manufacturing as a
whole, cost measures are computed as the cumulative percent change between the published and hypothetical
index values in 2007 and 1997. A cost inflation of 0.2, for example, represents a 20 percent increase in prices over
the decade. Two industries, petroleum products, and computer and electronic components, were included in the
overall manufacturing number but excluded from the charts. Petroleum products had cumulative input cost
inflation of 137 percent and bias-corrected inflation of 134 percent. Computer and peripherals had input costs
decline by 35 percent, 51 percent adjusted.
311
Table 1: Sources of Growth for U.S. Manufacturing Industries, 1997–20071
Gross
Output
(1)
MFP2
(2)
Capital3
(3)
Labor
(4)
Energy
(5)
Services
(6)
Purchased Materials
Domestic
Foreign
(7)
(8)
1. Manufacturing
2. Manufacturing excl. Computers and electronic products
1.18
0.46
1.30
0.69
0.13
0.11
-0.53
-0.47
-0.05
-0.05
0.22
0.13
-0.19
-0.23
0.28
0.28
3. Durable goods:
4.
Computer and electronic products
5.
Durable goods excl. Comp. & electr. products
2.00
7.35
0.77
2.02
6.82
0.95
0.17
0.25
0.15
-0.66
-1.11
-0.57
-0.05
-0.05
-0.05
0.30
1.05
0.12
-0.15
0.04
-0.22
0.37
0.35
0.38
6. Nondurable goods:
0.16
0.45
0.07
-0.37
-0.04
0.14
-0.25
0.17
1. Average annual rate for period shown. Column (1) is percent change. For each row, columns (2) through (8) are percentage points that sum across columns to (1).
2. MFP is multifactor productivity. 3. Includes Non-IT equipment, IT Capital, (computers and peripheral equipment, software, and communication equipment),
structures, and inventories
312
Table 2: Hypothetical Offshoring of "Obtanium"
t
t+1
t+2
t+3
Domestic supplier price
Domestic quantity sold
$10.00
100
$10.00
90
$10.00
80
$10.00
70
Chinese supplier price
Chinese quantity sold
$6.00
0
$6.00
10
$6.00
20
$6.00
30
$10.00
$9.60
$9.20
$8.80
Domestic input price index
Import input price index
Input index, as computed
100
─
100
100
100
100
100
100
100
100
100
100
True input price index
100
96
92
88
Average price paid for obtanium
Table 3: Foreign Offshoring and the Bias to U.S. Multifactor Productivity and Value Added, 1997-2007
Micro Evidence
Dev50,
Int30
Dev30,
Int15
(4)
(5)
(6)
1.05
0.52
1.12
0.58
1.08
0.54
1.16
0.61
1.87
6.33
0.89
1.64
5.91
0.70
1.73
6.13
0.76
1.67
6.05
0.71
1.77
6.18
0.81
0.45
0.45
0.36
0.40
0.38
0.42
3.04
0.94
2.82
0.86
2.31
0.44
2.50
0.59
2.39
0.48
2.61
0.68
9 Durable goods:
10
11
5.25
22.68
1.74
4.86
21.12
1.58
4.19
19.73
1.05
4.44
20.44
1.22
4.27
20.17
1.07
4.57
20.61
1.34
12 Nondurable goods:
0.07
0.08
-0.23
-0.10
-0.15
-0.03
Simulation:
Full Sample Switching
Baseline
IPP=PPI
(1)
(2)
(3)
Multifactor Productivity:
1 Manufacturing
2 Manuf.excl. Comp. & electronic products
1.30
0.69
1.23
0.67
3 Durable goods:
4
5
2.02
6.82
0.95
6 Nondurable goods:
Value Added:
7 Manufacturing
8 Manuf.excl. Comp. & electronic products
NOTE: Figures represent average annual percent growth in real value-added. For "IPP=PPI,” imported materials are deflated with domestic
deflators. For “full sample,” estimates are adjusted with product-level discounts from the entire IPP micro data sample. For “switchers,” the
import discount is based on a sample where importers appeared to shift from suppliers in advanced counties to ones in developing or
intermediate countries. “50/30” assumes discounts of 50 percent for developing countries and 30 percent for intermediate countries, while
“30/15” assumes discounts of 30 percent and 15 percent respectively.
313
Producing an Input Price Index
William Alterman
Bureau of Labor Statistics
November 2009
315
This paper is designed to address the need, and especially the feasibility, of producing an input
price index at the U.S. Bureau of Labor Statistics (BLS). These price indexes would serve to
provide more accurate estimates of several key indicators of the state of the U.S. economy,
including gross domestic product (GDP), productivity, and inflation.
The current interest in these types of price indexes arose over concerns that the BLS does not
adequately measure shifts in prices resulting from offshoring (or its corollary, onshoring) in its
industrial price programs. The BLS has three indexes that cover the production of goods, the
International Price Program’s (IPP’s) import price index (MPI) and export price index (XPI), and
the producer price index (PPI). The MPI only covers goods that are being imported, the XPI
only covers the export of goods, and the PPI only covers goods and services that are produced
domestically. Thus, a good that had been domestically produced and repriced by the PPI, and
has had its production sent overseas, will no longer be tracked in the PPI. Correspondingly, the
MPI index will not begin to price that particular item until after it has become an import.
Therefore neither program will directly show the price change that occurs when the item goes
from domestic production to foreign (or vice versa).
An example of how the BLS constructs an import price index and a producer price index will
help to illuminate the problem. Let us look at how both indexes might reflect price changes in the
manufacturing of furniture. Below I’ve constructed a table showing prices for four different
chairs. All chairs that are being produced domestically sell for $10, while all imported chairs sell
for $5. Chair A is only produced domestically, while Chair D is only imported. During the year,
the remaining two chairs shift from domestic production to being imported, Chair B in March
and Chair C in May.
The PPI only tracks Chair A for the entire period, and Chairs B and C for the months that they
are domestically produced. The MPI only tracks chair D for the entire period, and chairs B and
C only for the months they are imported. Thus, the PPI and the MPI for chairs would both
reflect no change during the entire reference period.
One suggested option was to combine the two indexes. However, since the indexes themselves
are always unchanged, no amount of recombining or reweighting will produce anything other
than an unchanged series. The only way to construct a price index that would show the price
decline associated from the offshoring of chairs B and C would be to construct a price index that
would directly track the price changes of items as they move from domestic to foreign and vice
versa. This is not possible under the methodology (and concepts) currently in use in the BLS’s
two industrial price programs.169 The PPI does currently construct output price indexes for
wholesalers and retailers, which presumably includes data on both imported and domestically
produced goods. However, these indexes are only gross margin indexes, and only represent the
difference between their selling price of a good and the acquisition price for that same item. The
data collected does not lend itself to delineating import goods from domestic goods.
169
Note that the consumer price index is designed to pick up these price changes, and is reflected in prices paid by
domestic consumers. In addition, the bureau has conducted a preliminary analysis of PPI data that provides some
evidence that prices from domestic producers are influenced by the degree of import penetration in their industry.
316
Jan-09
Feb-09
Mar-09
Apr-09
May-09
Jun-09
$10
$10
$10
$10
$5
$5
$5
$5
$10
$10
$5
$5
Chair A
Domestic
$10
$10
Chair B
Domestic
$10
$10
Chair B
Imported
Chair C
Domestic
Chair C
Imported
Chair D
Imported
$10
$10
$5
$5
$5
$5
$5
$5
PPI
100
100
100
100
100
100
MPI
100
100
100
100
100
100
Combined index
100
100
100
100
100
100
Input index
100
100
85.7
85.7
71.4
71.4
Although the BLS was aware of the potential data gaps between XPI, MPI, and PPI, it appeared
that shifts over time between domestic and foreign production have been gradual enough that it
was not evident that the limitation of the indexes was seriously biasing estimates of productivity
as well as GDP figures and other components of the National Accounts. The potential
shortcomings in the BLS indexes, however, were highlighted in an article in the summer of 2007
in Business Week, and by a study funded by the Sloan Foundation (Mandel 2007; Houseman
2009a). Presumably this potential gap in BLS data becomes more serious as the proportion of
the U.S. economy tied into the global economy has grown, especially in conjunction with the
perception that U.S. jobs are being lost to foreign competition.
In order to address this limitation, the BLS needs to develop an entirely new set of “input” price
indexes, which would price goods and services that are inputs into a domestic company’s
production function. Indeed, the BLS itself recognized the need for this type of series over 30
years ago when the old “wholesale price index” was transformed into the more comprehensive
and systematic output-based producer price indexes.
This paper will review both the concepts and uses of an input price index, as well as assess
additional evidence centering on the need for these data. In addition, the paper will also focus on
the practical aspects and limitations of attempting to produce such an index. This will include
surveying the data sources necessary for drawing a sample of establishments and items to
reprice, evaluating possible sources for appropriate weights in an input price index, determining
a proper index estimation formula, and verifying the publication structure necessary to support
the different uses of these series.
317
WHY AN INPUT PRICE INDEX IS IMPORTANT
As mentioned, an accurate estimate of the prices paid for inputs of both goods and services is
crucial to a number of broad and critical measures of the economy. This includes estimates of
GDP, inflation, and productivity. For example, in order to properly estimate GDP by industry,
(constructed by the Bureau of Economic Analysis [BEA]) and industry productivity estimates
(constructed by the BLS), these agencies must properly account for input costs. Although these
data are available on a current dollar basis (though sometimes with a considerable lag), in order
to estimate their “real” (that is, inflation-adjusted) values, they need to be deflated by changes in
price levels. However, the appropriate price measures paralleling these input values are not
currently being produced by the BLS. Consequently, the BEA and the BLS must make use of
whatever price data are available. Generally this has required the agencies to make use of the
PPI output price indexes and/or the IPP import price indexes. It has been speculated that using
these next best sources may lead to significant mismeasurements in the economy. For example,
the Business Week article (Mandel 2007) estimates that the increase in real GDP from 2003 to
mid-2007 may have been overestimated by $66 billion. This article focuses on import prices not
picking up price changes when a good goes from being domestically produced to being
imported. It summarizes the example of the furniture industry and highlights the apparent
contradictory behavior of consumer prices for furniture; those prices have been decreasing while
the indexes for both domestic producer prices and import prices for this category have both been
increasing.
Equally important, the article also infers that the lack of an input price index may lead to a
significant overestimate of productivity in U.S. industry. A rise in a nation’s productivity is
considered the key factor in an economy’s ability to improve its standard of living. It is further
assumed that increases in real hourly earnings are often tied to gains in productivity. If, in fact,
GDP and productivity are being overestimated, this implies that the gains from trade (that is, the
terms of trade) are being underestimated, and that in real terms the value of imports is greater
than currently measured.
RECENT WORK
A growing body of literature has now examined the increasing role of imports in intermediate
inputs in the U.S. economy, as well as concerns associated with the methodology in constructing
U.S. estimates of GDP and productivity. Kurz and Lengermann (2008) note that foreign inputs
accounted for one-third of growth in the manufacturing sector between 1997 and 2005, while
Yuskavage, Strassner, and Medeiros (2008) estimate that from 1997 to 2006 the import share of
intermediate inputs increased from 13.5 to 20.0 percent. Feenstra, Reinsdorf, and Slaughter
(2008) attribute a substantial portion of the apparent acceleration in productivity gains after 1995
to gains in the terms of trade and tariff reductions. Nakamura and Steinsson (2008) find
limitation in the import and export price indexes associated with “product replacement bias.”
Finally, Houseman (2009b) states, “The measurement problem has broad implications not only
for various aggregate and industry statistics, but also for the research that relies on them.
Although the growth of imports from developing countries has spurred great interest in academic
318
and policy circles about their effects on the U.S. economy and its workers, credible research into
these issues cannot be conducted without accurate data on real import values.”
Additional Evidence
In order to prove the need for a set of input price indexes that incorporates both domestic and
foreign sourcing, I analyzed the most recent available data on the role of imports in domestic
supply. In analyzing the data from the BEA, I estimate that not only has the contribution of
imports to intermediate inputs in the United States increased, but that it increased at a faster rate
during the past decade. In 1975, imports represented less than 7 percent of inputs into
manufacturing. By 2007 that figure had climbed to almost 28 percent (see Figure 1). Equally
important, between 1997 and 2007, the percent of imports in inputs increased by an average of
over 0.4 percent a year, while in the prior decade, the percent had increased by less than 0.25
percent a year. This point is critical because it infers that there is an acceleration in companies
shifting their products from domestic sourcing to foreign sourcing, making the need for
additional data more critical. In addition, if the rate of change was consistent over time, it might
have been easier to model a “discount” factor to apply to import prices in order to adjust for this
shift.
Indeed, the speed of globalization is perhaps happening so quickly that the ability of traditional
measures to capture these shifts has become increasingly problematic. For example, the
household wood furniture manufacturing industry—the industry highlighted in the Business
Week article—recorded a dramatic increase in the value of imports during the past decade,
jumping from $13.2 billion in 1999 to $27.0 billion in 2007. Despite this increase, in 2006 the
preliminary estimate from the Annual Survey of Manufactures for the household wood furniture
sector recorded an increase in value of domestic production, up to $13.5 billion. However, when
the final figures were revised the following year, the number was adjusted substantially
downward to only $8.6 billion. This may be due in part to the difficulty of properly (and in a
timely manner) coding companies to the correct NAICS (North American Industry Classification
System) number when they shift from being a manufacturer to being essentially a wholesaler.
Limitations
It is important to point out that the construction of an input price index will not alleviate directly
the potential mismeasurement issues associated with an import price index. This is important to
note because GDP can be estimated using two different methods: It can be constructed by
calculating the total of final sales in the U.S. economy and subtracting imports (the familiar Y =
C + I + G + X – M), as well as by the value-added approach, where the total value added of each
industry is aggregated (Y = VAj, where VAj = Sales for industry j – Purchases of materials and
supplies by industry j. The current methodology in the United States focuses on the former.
In order to understand why the BLS cannot construct an import price index that directly registers
these price changes, it helps to review the current methodology. The procedure for producing
import price indexes starts out with a very robust frame from which to draw a sample. It
includes nearly the entire set of transactions of all merchandise brought through U.S. Customs
and into the United States. It categorizes it by individual shipments, product categories, and of
319
course, companies. A sample of specific companies and the items they imported is then drawn
from this frame, and the BLS attempts to collect prices on a monthly basis for these items. Note,
however, that the sample only consists of goods that are already being imported. It is not
practical to ascertain from an importer (who in many cases may only be an intermediary) if in the
past he sourced an item domestically. It would also be hard to get the reverse, asking an
importer who no longer imports if the sampled good is now produced domestically, and if so,
what is the price. Presumably, constructing an input price index may potentially provide some
indication of the magnitude of any differences in price trends being missed by import prices or
producer prices as sourcing shifts from one to the other. This might be possible if, as the pricing
data was being collected, the respondent was able to take note of whether the item was bought
domestically or from a foreign source. From a practical standpoint, however, it is not clear if this
information could be successfully incorporated into the index production process.
It should also be noted that an input price index will not alleviate problems arising from
previously in-house-produced goods and services that are now being outsourced (either
domestically or to a foreign source). This, too, is considered a growing phenomenon, but unless
data on prices associated with the in-house cost of producing an item can be directly compared
with the outsourced price, it is not clear how the BLS could evaluate shifts in prices associated
with this phenomenon.
BLS AND INPUT PRICE INDEXES
The seminal 1961 report of the National Bureau of Economic Research Price Statistics Review
Committee, the so-called Stigler Report, made a number of recommendations surrounding the
wholesale price indexes, which was the name of the industrial price series then being produced
by the BLS. One of the recommendations was that the BLS should rely on buyers’ prices rather
than sellers’. A second recommendation was for the creation of a set of conceptually rigorous
input and output price indexes. The report also included an empirical study, which concluded
that buyers’ prices were more likely than list prices to accurately reflect prices of actual
transactions.
Buyers’ Prices
Prior to the Stigler Report, the PPI had evaluated the use of buyers’ prices. In 1942, the PPI
conducted a study of buyers’ prices for eight selected items of steel mill products for six time
periods and compared them to list prices. The results of the study showed that the buyers’ prices
moved differently than list prices for short periods of time, but longer-term list and invoice prices
were comparable. Experiences with the study showed that purchases of an item by an individual
company included many different transaction terms and detailed specifications.
In response to the Stigler Report and subsequent reports, the BLS commissioner as well as others
expressed concerns that the cost of collecting buyers’ prices would outweigh the likely benefits
due to potential problems such as buyers’ prices from an invoice sometimes not reflecting real
transaction prices, difficulties capturing retroactive price adjustments based on cumulative
volume, and financial assistance given by sellers to buyers for advertising and other expenses.
320
The BLS did agree that the project had merit on a selective basis to allow analysis of price trends
in industries where transaction pricing was difficult.
A more detailed study looking into the advantages of buyers’ prices was published in Stigler and
Kindahl (1969), which pointed out the differences in price trends between buyers’ and sellers’
prices. As much of the concern with the then BLS wholesale price index focused on the use (or
potential misuse) of so-called list prices, the BLS economists worked with the sellers who were
participating in the price survey to encourage the reporting of actual transaction prices and made
substantial progress in some industries in improving the quality of the received prices. In
addition, the PPI also began the process of evaluating specific products where buyers’ prices
should be collected due to the unavailability of transaction prices from sellers. As a result of this
study, in January 1972 the PPI began publishing a commodity index for aluminum ingot using
buyers’ prices from a judgmentally selected sample of reporters.
Building on this work, in 1974 the PPI attempted a systematic sampling approach to obtaining
buyers’ prices. This project was undertaken with the goal of determining the feasibility and cost
of collecting prices directly from buyers in order to either calculate price indexes or evaluate the
quality of the transaction prices being reported by sellers. The project identified highly
weighted products where either sellers refused to provide transaction data or the quality of
current transaction data was questionable, and where there were homogeneous products
frequently purchased by buyers in consistent quantities. The project focused on titanium forgings
because the PPI was able to create an unrefined frame and document the typical transaction
characteristics of buyers in this product area. After significant resources were spent on this
project, pricing issues remained, and a process had not been defined to refine and systematically
sample from the frame. As a result, the project was dropped and the program reverted to
obtaining good transaction prices from sellers even in these more difficult cases. No further
work was done on buyers’ prices, and in 1980, when indexes calculated using sellers’ transaction
prices were introduced from the systematic sample for the primary aluminum industry output
index, the buyers’ price commodity index for aluminum ingot was dropped.
Input/Output Indexes
In response to the Stigler Report, the PPI also began examining approaches to creating input and
output (price) indexes for industries. For example, in the early 1960s the PPI built output
Industry-Sector Price Indexes (ISPI) for some industries by combining the judgmentally sampled
data collected for the commodity indexes using different classification structures and weighting.
In the mid 1970s, however, the PPI began a comprehensive revision in order to plan and
implement many improvements that had been recommended over the years, including those in
the Stigler Report. The long-term goal of the revision was to expand the PPI’s coverage to every
industry in the private economy and to publish a system of price indexes that included
•
industry output indexes,
•
industry input indexes,
•
detailed commodity indexes, and
•
industry based stage of processing indexes.
321
In the late 1970s the PPI began systematically sampling industries, and in 1980 began
introducing industry output indexes on a regular basis.170 Throughout the years, the PPI
continued to expand the number of industry output indexes and now covers 82 percent of all inscope production.
As an attempt to fulfill the recommendations of the Stigler report, and as a component of its
stage of processing indexes, the BLS did publish a set of input price indexes from 1988 to 2003.
These indexes were calculated by reweighting output prices using input weights, which allowed
the use of output price indexes at a great level of detail. However, these indexes did not include
imports, nor did they directly account for substitution from a buyer’s perspective. Thus, they
assumed that sellers’ prices are a good proxy for buyers’ prices, and that prices for imports and
domestic production move similarly. These series were discontinued in 2003, but the BEA and
the BLS still used them as a method for constructing input price indexes where necessary (see
Table 1).
Note that the BLS does have extensive experience with constructing an input price index,
because both the import price and consumer price indexes are constructed from buyers’ prices.
CURRENT USES AND USERS OF THE DATA
The fundamental question facing the BLS, of course, is, can the BLS produce an input price
series that will meet the needs of its primary users? To answer this question, one must first delve
into the intricacies of the construction of the outputs of the two primary potential users of these
data, the Office of Productivity and Technology (OPT) at the BLS, and the Industry Sector
Division of the BEA.
The BLS
We will start with the OPT, which publishes two types of productivity measures: 1) labor
productivity, or output per hour of labor; and 2) multifactor productivity, or output per unit of
combined inputs. Labor productivity indexes and multifactor productivity indexes are produced
in two different divisions in the BLS.
Labor productivity
Measures of labor productivity are produced in two divisions of the OPT: the Division of Major
Sector Productivity (DMSP) and the Division of Industry Productivity Studies (DIPS). The
estimates of labor productivity (and unit labor costs) for major sectors are published quarterly,
while estimates for industries are published annually. Labor productivity estimates do not
explicitly measure shifts in the quantity (or constant dollar value) of material inputs, and
therefore do not require estimates of the changes in the prices of those inputs, be they
domestically sourced or imported. Note that outputs are adjusted for inflation.
170
While the practical work focused on an output price index, work did proceed on the theory of an input price
index, culminating in a BLS working paper by Robert Archibald in 1975.
322
Multifactor productivity
Multifactor productivity measures are also produced in both the DMSP and DIPS. The DMSP
publishes, albeit with little detail, multifactor productivity estimates for the private business and
private nonfarm business sectors of the economy. These series represent 77 percent of U.S.
GDP. In calculating these series, outputs are measured on a value-added basis, which are then
compared to just two inputs, capital and labor. The value of material inputs is excluded from
these calculations. However, staff uses detailed price indexes to deflate capital expenditures.
Physical capital, as measured by the DMSP, consists of 42 types of equipment and software, 21
types of nonresidential structures, 9 types of residential capital, inventories (manufacturing available
for 3 stages of fabrication), and land. Deflation of each capital expenditure category is actually
done at the detailed 5- or 6-digit input-output (I-O) level. (The actual derivation of value-added
by sector entails adjusting the value of inputs to account for changes in prices. This work,
however, is done at the BEA.)
The DMSP also publishes annual multifactor productivity measures for total manufacturing and
18 broad 3-digit NAICS manufacturing industries, comparing sectoral output (total output
excluding intraindustry or intrasector transactions) to a broad set of inputs, including capital,
labor, energy, materials and business services (KLEMS) inputs. (Note that on a value-added
basis, manufacturing represented 12 percent of GDP in 2007.) In the manufacturing sector of the
economy and in individual industries, intermediate purchases constitute the largest component of
inputs. The nominal dollar and constant dollar values of energy, materials, and services used by
the DMSP are derived by the BEA.
DIPS publishes more detailed annual multifactor productivity measures for 86 4-digit NAICS
manufacturing industries, plus air transportation and line-haul railroads. These productivity
measures also compare industry sectoral output to a broad set of combined inputs. DIPS
publishes estimates of intermediate purchases, capital, and labor for each of the detailed
manufacturing industries. The index of intermediate purchases for each industry is constructed
by combining separate quantities (or constant dollar costs) of electricity, fuels, materials, and
purchased services. In order to deflate nominal dollar cost inputs for each industry, weighted
deflators for materials and services are calculated by combining detailed price indexes using
weights derived from the cost of commodities consumed by each industry, as shown in the
detailed benchmark I-O tables produced by the BEA. I-O commodities from the benchmark I-O
tables generally relate to the primary products of 6-digit NAICS industries, or occasionally a
combination of industries. For materials commodities that are heavily imported, DIPS combines
PPIs and import price indexes using weights from the BEA’s import matrix. DIPS also uses
PPIs to create weighted deflators to deflate annual fuels purchases of each industry and capital
expenditures. Price deflators for each equipment asset category are constructed by combining
detailed PPIs with weights from the BEA capital flow tables at the roughly 6-digit level. For the
DIPS detailed manufacturing industry measures, physical capital consists of 25 categories of
equipment, two categories of structures, three categories of inventories, and land.
Note that the BLS makes use of product-specific data in constructing deflators for a set of input
price indexes for a given industry’s material costs. Ideally, an input price index would be
industry-specific, but that may prove cost-prohibitive.
323
Since industry MFP calculations are based on annual data, the nominal input values are adjusted
by annual PPIs (average of 12 monthly price indexes).
The BEA
The Industry Sector Division at the BEA is responsible for producing the annual industry
accounts and the benchmark input-output accounts. These accounts, which shed critical light on
the relationships between U.S. industries, take a value-added approach to and are consistent with
the BEA’s flagship GDP estimates. Although the BEA does not publish detailed annual real
input-output estimates, they do publish annual price and quantity indexes for 65 detailed
industries, including 19 manufacturing industries, which do require data on the real value of
inputs.
As in the work at the BLS, the BEA attempts to make its adjustments at the most detailed level
possible. For example, at the BEA, the effort to construct updated values for intermediate inputs
of goods and services entails making adjustment to approximately 3,500 different items, of
which roughly 2,300 represent categories of goods. Ideally, and like the BLS, the BEA would
like input price indexes by industry for each of the 1,179 6-digit NAICS level of detail. In
practice, since the cost of producing that many separate price indexes could be prohibitive, the
BEA, like the BLS, would accept a set of product-based input price indexes. In addition, at a
minimum, category definitions should be consistent with the 12 expense categories recently
added to the Census Bureau’s ASM forms (most of which are services inputs). While the BEA
currently only produces annual estimates of GDP-by-industry, there has been growing interest in
providing these estimates on a quarterly basis.
In sum, although superficially the level of publication required to produce the currently
published set of economic data is comparatively high, in actuality the detail necessary to
properly support these estimates may be considerably more disaggregated.
STEPS TO PRODUCE AN INPUT PRICE INDEX
While there is little dispute over the potential advantages of adding an input price index to the
family of price indexes produced by BLS, there is the fundamental question of both feasibility
and cost of producing a usable and comprehensive set of indexes.
Developing a Sample
From a practical standpoint, the first and perhaps the biggest hurdle in developing an input price
index is developing a frame from which to draw a sample of establishments; the bureau does not
currently have access to data on the expenses and purchases of individual companies necessary
to produce a representative sample. Without these data, a BLS field agent attempting to initiate a
respondent into a survey would have no information on what that establishment buys in order to
produce its outputs. While, in theory, the establishment might be able to supply these data, in
practice it is expected that this type of data collection would be very problematic given the
voluntary nature of BLS programs.
324
All is not lost, however, as the Census Bureau does collect detailed data on purchases by
individual establishment. In particular, in the Economic Census, which the agency conducts
every five years, all manufacturing firms are asked to include detailed data (by 10-digit NAICS
code) on their cost of materials, parts, and supplies consumed in the reference year. The most
recent data available cover the calendar 2007 Economic Census and became available in mid2009. The dataset includes information for 340,000 manufacturing establishments in the United
States, and the Census Bureau records the total cost of materials purchased by these
establishments as approximately $2.5 billion in calendar 2007. Table 2 is part of the collection
form for the MC-33702 Manufacturing, Household Furniture and Wood Housings sector, where
establishments report on their material costs. In addition, Table 3 has an example of the type of
data that is publically available from the census. For NAICS 333111, one can find data on the
number of companies and their total purchases and expenses, as well as an indication of their
relative size. Table 4 shows data on cost of materials by type of material for that same industry,
while Table 5 reflects the total purchases for all manufacturing industries of a given commodity.
In addition, the less comprehensive but timelier Annual Survey of Manufacturing, which is based
on a sample of 50,000 manufacturing establishments, includes a limited amount of data on
purchases, providing one category for total cost of materials, parts, containers, packaging, etc.
One shortcoming of these surveys is that, while data on capital expenditures is also collected, it
is only split three ways: 1) motor vehicles, 2) computers, and 3) other. Another potential
shortcoming of this source of data is its timeliness, or lack thereof. Since the detailed data is
only collected once every five years, it may be that by the time the BLS is able to draw a sample
and initiate these establishments into a market basket, the establishments and/or the products that
they buy may be out of date and no longer reflective of their current market.
Although much of the focus has been on the manufacturing sector, it should be recalled that the
manufacturing component only accounts for a small and shrinking sector of the economy;
services represent nearly two-thirds of GDP. The amount of detailed cost data collected by the
census for the service industry surveys is more limited. In general the collection forms include
some detailed data on purchased services, but only limited data on purchased equipment and
materials.171 Interestingly, while the census collects very little detailed data on material costs in
the noncensus years for manufacturing industries, the level of detailed data collected for the cost
of business services, though limited, is roughly the same, whether it is for an annual survey or
the every five-year census. In general, the surveys break out the purchases of business services
into five categories: computer services, communication services, advertising and related
services, professional and technical services, and repair and maintenance services.
Due to the more detailed cost of materials data available for the manufacturing sector, much of
the current assessment of a potential sampling frame has focused on this sector. Unfortunately,
because many of the datasets at the Census Bureau have data that has been commingled with
171
For example, in contrast to the forms for the furniture manufacturing industries, the collection form for the
parallel furniture wholesale (Table 6: WH-42305) sector does not provide the same level of detail on material costs,
while the collection form for the retail furniture industry (Table 7: RE-44201) does not collect any information on
the cost of materials.
325
federal tax return information data from the Internal Revenue Service, getting access to the
necessary data has been somewhat problematic. Work has continued for several years on what is
referred to as companion legislation.172 Regardless, BLS staff have recently been able to access
these data at the Census Research Data Center in Suitland, Maryland, and have begun the
process of assessing the utility of using these data to draw samples that would permit the
publication of input price indexes for the 471 6-digit NAICS manufacturing industries. One
concern is that a large percentage of the cost of materials purchased is in a miscellaneous
purchase category.
Assuming BLS is able to use the census data, this would allow the BLS, using establishment
sampling methodologies with which the bureau is already quite familiar, to construct a sample of
establishments, and detailed product areas within the given establishment, that the bureau would
need to collect the necessary pricing data. The selection of the actual item that the bureau would
need to reprice on a periodic basis would normally be done by a BLS field economist during a
so-called initiation visit to the establishment. This procedure is one that is already done by
staffers when collecting data for the bureau’s PPI and IPP industrial price programs, and
involves a number of trade-offs. Ideally the selection would be based on a probability
proportionate to how much of a given item a company purchases within the selected category.
Thus, if a company buys a certain amount of varying types of steel, the field economist, using
data hopefully supplied by the respondent, would be able to select a specific steel product on
which the BLS would attempt to collect data. In practice, however, these procedures would
likely have to take into account the fact that the selected item may not be purchased on a regular
basis, or the respondent may not have any data available on how much of each different type of
steel the company purchased in a given period. Since the BLS already has experience with these
types of issues in its current programs, developing an appropriate fallback procedure does not
necessarily present a problem. However, it does lead in to what is perhaps the key issue, which
is the ability of the program to reprice the same item month after month, quarter after quarter or
year after year, from the same source.
Pricing
Maintaining a constant set of items to reprice over time may prove to be the most intractable
barrier to constructing a comprehensive set of input price indexes. While on the output side,
companies tend to ship their goods (or offer their services) every month, it is not clear if they buy
the same item on a regular basis, especially for capital equipment such as computers. This may
place a heavier burden on the imputation method chosen for valuing prices in missing periods.173
Alternatively, the BLS may have to use an altogether different approach, such as combining
prices from different respondents (in cases where the item specifications are identical). A related
question is how to handle changes in the pricing specifications. For example, if the product is
the same but the supplier is different, do we continue to price it as the same item? What is our
172
Legislation to modify the IRS tax code was proposed by the last Administration, with interagency support from
the Departments of Treasury, Commerce, and Labor, in 2002–2003 and in 2008. Conversations have begun on the
development of an official Obama Administration proposal, with the Office of the Under Secretary of Commerce for
Economic Affairs taking the lead on this effort.
173
In constructing a sample for the import price index, the International Price Program has the advantage of
accessing the universe of import transactions from the Customs Service, which allows for drawing a sample only of
those items and importers who trade consistently over the course of a year.
326
general approach toward quality adjustment when a buyer switches products and/or suppliers?
That is, in an ideal situation where we can get the exact information that we desire, what would
we ask for? What are the acceptable fallbacks if we can’t obtain the desired information? What
if, in fact, the buyer uses multiple suppliers? Do we select a specific supplier or use some sort of
average? If we select one, how and when do we switch to a price from a different supplier?
Should the price include or exclude transportation costs? If other services are bundled with the
product (e.g., installation), how do we handle those situations? Do we want to include
government purchases? If so, how would we sample for them since they wouldn’t be included in
data at census? How do we coordinate requests for buyers prices with requests for sellers prices
within the same firm?
In order to answer these and similar questions, the BLS will most likely need to make some
effort to collect information from a sample of representative companies. A final decision on
some of these issues will probably entail balancing the requirements of a price index with the
reality of the agency’s sometimes limited ability to collect data voluntarily from private industry.
Estimation Formula
With one exception, compared to the questions associated with sampling and repricing, the
issues surrounding the estimation formula are comparatively easy. Weights can either be derived
from the sampling frame, from the respondents themselves, or from some combination thereof.
One concern with using the weights derived from the sampling frame is the age of the data.
Since the detailed data are collected only once every five years, the data may be out of date by
the time they are actually used in the calculation of the indexes. A comparison of these values
from one census to the next may shed light on the volatility of these figures.
There are various questions associated with the actual formula to use, such as using an arithmetic
formula versus a geomeans formula, but they do not present intractable barriers. One interesting
aspect of the formula relates to theoretical differences between the price index formula for the
output from a production function versus the index formula for the price index for inputs into a
production function. The theory assumes that a firm will attempt to maximize profits by
minimizing costs while maximizing revenue. On the output side, theory tells us that an
establishment will attempt to shift sales to its goods or services that over time are becoming
relatively more expensive compared to its other outputs. In contrast, the firm would attempt to
shift costs toward its expense categories that are becoming relatively cheaper. Consequently, a
price index of firms’ outputs would tend to show at least no decline in the relative quantity of the
more expensive goods being sold, while on the cost side, the index should in theory reflect at
least no increase in the goods or services that are more expensive. Interestingly, these
assumptions are based on partial equilibrium models where the model is only looking at one side
of the equation. But, of course, one establishment’s sales are another establishment’s purchases,
and in a general equilibrium model, there is no a priori theory of exactly what constitutes the
correct direction of substitution.174
The one notable difficulty in estimating these indexes relates to how one goes about constructing
industry-specific price indexes. While a product-based input price index would use every
174
For further elucidation, see Kim and To (2009)
327
establishment’s purchases of a specific good (or service), an industry-specific input price index
would only use goods or services purchased by establishments in that specific industry. For
example, presumably all establishments must purchase energy, be it electricity, gas, petroleum
products, etc. Would the BLS attempt to calculate a separate energy index for each industry, or
would it combine all energy data into one generic input energy index? The answer may
ultimately be decided on practical grounds, (i.e., do we have enough data for separate energy
series, or do each of the different energy series trend nearly the same?) Of course, a proxy for an
industry-specific input price index could be constructed using individual product-level price
indexes, but aggregating them using the proportions appropriate for a particular industry’s
purchasing patterns.
Next Step: Reality Check
As previously mentioned, the preliminary step in this effort to produce an input price index is to
develop a set of questions for a limited set of respondents who in the past have proven
cooperative to the BLS industrial price programs. The questions would be designed not to
collect specific pricing data from the establishments, but to provide the BLS with some insight
into how respondents are likely to react to any such data collection requests. For example, some
companies have refused to provide BLS with data, citing confidentiality concerns. Thus, one
question might be designed to ascertain whether companies would be more likely, less likely, or
equally as likely to supply data on costs as they are willing to supply output or import or export
price data. Another basic question to ask respondents would be to ascertain if they even keep
good data on the cost of their purchases, and if so, what the periodicity of their purchasing is.
Developing a Pilot
A longer-term effort to produce input price indexes can be broken down into four phases, based
on availability of data. This effort will require additional approvals and funding as well. The
four phases include input indexes covering 1) manufacturers material costs, 2) manufacturers
capital equipment costs, 3) manufacturers business services costs, and 4) service industries
material, capital equipment, and business services costs.
Ideally, each phase would start with a pilot prior to going into production. For each pilot, BLS
will conduct research and develop the methodology, procedures, and systems associated with
each of the following activities:
•
Obtain permission from the OMB.
•
Select a set of industries for the pilot.
•
Evaluate the data sources that are available for a sampling frame. Due to the potential for
detailed cost data from the Census of Manufactures, the first phase would focus on input
indexes of cost of materials for manufacturing industries.
•
Develop the collection materials and procedures and train staff.
•
Select a sample of establishments for the pilot.
•
Conduct the pilot test and evaluate the results.
328
•
Evaluate the feasibility of producing an input price index for the given phase and develop
the requirements for producing an input price index, including publication goals, required
sample size, expected burden, and estimated resources and timeframe for publication.
•
Based on available resources, develop and maintain a production set of indexes for that
particular set of input indexes.
Cost
Assuming the methodological and data collection issues can be resolved, and assuming the BLS
is able to collect the necessary data from respondents, there remains the question of the cost of
developing and maintaining these new indexes. On the one hand, the collection, review, and
verification of data for inclusion in price indexes still has a significant labor-intensive
component, usually requiring a substantial level of expertise in economics and/or statistics. On
the other hand, a significant (but unknown) amount of the necessary resources, both in human
capital as well as data processing applications, may be shared with the bureau’s other industrial
price programs, the IPP and the PPI. Any bureau effort to produce an input price index past the
research phase would require resources sufficient to cover collecting approximately 15,000 items
and publishing approximately 600 6-digit material codes (which are similar to NAICS codes).
The process for developing these series would extend over several years. Extending the set of
indexes to cover the three additional phases would entail an additional annual cost.
329
Table 1
330
Table 2
2007 Economic Census: MC-33702 Manufacturing,
Household Furniture and Wood Housings
331
Table 3
NAICS 333111
Farm Machinery and Equipment Manufacturing
Companies Establishments
with 100
employees or
More
Total Value of
Shipments
($1,000)
Total Capital
expenditures
($1,000)
Total Cost of,
Purchased
materials
($1,000)
1,079
$21,181,238
$348,399
$9,903,172
104
Table 4
NAICS 333111
Farm Machinery and Equipment Manufacturing
(Cost of Materials)
Material Code
Delivered cost
($1,000)
Description
971000
Materials, ingredients, containers, and supplies, nsk
970099
All other materials/components/parts/containers/supplies
967,152
33000019
Engines (diesel/semidiesel/gasoline/carburetor-type/etc.) &
parts
680,000
33000067
Fluid power products, hydraulic and pneumatic
607,834
33100022
Steel sheet and strip (including tinplate)
586,586
33200046
Other fabricated metal products (excl. Forgings/castings etc.)
504,553
32621103
Pneumatic tires and inner tubes
389,781
33635003
Transmissions and parts
288,496
33100025
Steel struct shapes & sheet piling (excl castings/forgings/etc.)
286,917
33361200
Mechanical speed changers, gears, & ind. high-speed drives
281,122
33120092
All other steel shapes/forms (exc. castings/forgings/etc.)
280,209
33151001
Iron and steel castings (rough and semifinished)
268,893
33632200
Engine electrical equip. (incl. spark plugs/magnetos/etc.)
226,547
332
2,718,394
Table 5
Expenditures on Fluid Power
products (Material Code 33000067) by Industry
Delivered
cost
($1,000)
NAICS
Code
Description
333111
Farm machinery and equipment manufacturing
607,834
333112
Lawn and garden equipment manufacturing
218,356
333319
Other commercial and service industry machinery manufacturing
422,091
333512
Machine tool (metal cutting types) manufacturing
66,118
333513
Machine tool (metal forming types) manufacturing
43,371
333516
Rolling mill machinery and equipment manufacturing
12,355
333518
Other metalworking machinery manufacturing
29,007
333611
Turbine and turbine generator set units manufacturing
333618
Other engine equipment manufacturing
284,283
336312
Gasoline engine and engine parts manufacturing
268,662
336330
Motor vehicle steering and suspension parts
89,222
336340
Motor vehicle brake system manufacturing
47,397
336350
Motor vehicle transmission and power train parts manufacturing
237,914
336399
All other motor vehicle parts manufacturing
405,854
333
4,687
Table 6
2007 Economic Census: WH-42305 Wholesale, Furniture and Home Furnishings
Table 7
2007 Economic Census: RT-44201, Retail Furniture Stores
334
Figure 1
Imports as a Percent of Domestic Supply
Manufacturing Sector
30.0%
25.0%
20.0%
15.0%
10.0%
5.0%
"Import Penetration Ratio"
335
20
06
20
04
20
02
20
00
19
98
19
96
19
94
19
92
19
90
19
88
19
86
19
84
19
82
19
80
19
78
19
76
0.0%
References
Archibald, Robert B. 1975. “On the Theory of Industrial Price Measurement: Input Price
Indexes.” BLS Working Paper No. 44, May. Washington, DC: Bureau of Labor Statistics.
Houseman, Susan N. 2009a. “Measuring Offshore Outsourcing and Offshoring: Problems for
Economic Statistics.” Employment Research 16(1): 1–3.
Houseman, Susan N. 2009b. “Outsourcing and Offshoring: Problems for Price and Productivity
Measurement and Implications for Labor Research.” Upjohn Institute Working Paper No.
06-130. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research.
Kim, Mina, and Ted To. 2009. “Input Prices Indices with International Trade.” Unpublished
manuscript. Bureau of Labor Statistics, Washington, DC.
Kurz, Christopher, and Paul Lengermann. 2008. “Outsourcing and U.S. Economic Growth: The
Role of Imported Intermediate Inputs.” Paper prepared for the 2008 World Congress on
National Accounts and Economic Performance Measures for Nations, held in Arlington,
VA, May 12–17.
Mandel, Michael. 2007. “The Real Cost of Offshoring.” Business Week, June 18.
Nakamura, Emi, and Jon Steinsson. 2009. “Lost in Transit: Product Replacement Bias and
Pricing to Market.” NBER Working Paper No. 15359. Cambridge, MA: National Bureau
of Economic Research.
Reinsdorf, Marshall B., Robert C Feenstra, and Matthew Slaughter. 2008. “Effects of Terms of
Trade Gains and Tariffs Changes on the Measurement of U.S. Productivity Growth.”
Paper prepared for the 2008 World Congress on National Accounts and Economic
Performance Measures for Nations, held in Arlington, VA, May 12–17.
Stigler, George, ed. 1961. The Price Statistics of the Federal Government. Report to the Office
of Statistical Standards. Bureau of the Budget. New York, National Bureau of Economic
Research.
Stigler, George, and James K. Kindahl. 1970. The Behavior of Industrial Prices. Washington,
DC: National Bureau of Economic Research.
Imported Inputs in the U.S. Economy: Insights from the Integrated Economic Accounts.”
Paper prepared for the 2008 World Congress on National Accounts and Economic
Performance Measures for Nations, held in Arlington, VA, May 12–17.
337
AND IMPORT USES: EVIDENCE OF BIASES
CHAIR: Carol Corrado (Conference Board)
Imported Inputs and Industry Contributions to Economic Growth: An Assessment
of Alternative Approaches, Erich Strassner, Robert Yuskavage
and Jennifer Lee (Bureau of Economic Analysis)
Evaluating Estimates of Materials Offshoring from U.S. Manufacturing,
Robert C. Feenstra (University of California-Davis) and J. Bradford Jensen (Georgetown
University and Peterson Institute for International Economics)
DISCUSSANT: William Milberg (New School for Social Research)
Errors from the “Proportionality Assumption” in the Measurement of Offshoring:
Application to German Labor Demand, Deborah Winkler and William Milberg (Schwartz Center
for Economic Policy Analysis, The New School for Social Research)
339
Imported Inputs and Industry Contributions to
Economic Growth:
An Assessment of Alternative Approaches
Erich H. Strassner, Robert E. Yuskavage, and Jennifer Lee
U.S. Department of Commerce
Bureau of Economic Analysis
Washington, DC
Paper prepared for the NAPA/Upjohn conference
“Measurement Issues Arising from the Growth of Globalization”
Washington, DC
November 6–7, 2009
341
Imported Inputs and Industry Contributions to Economic Growth:
An Assessment of Alternative Approaches
Over the past decade imports have become an increasingly important source of supply for both
U.S consumers and producers, partly due to changes in the relative prices of imported and
domestic goods. From 1997 to 2007, imports as a share of all goods and services consumed in
the United States increased from 18 percent to 23 percent. This aspect of globalization has
affected the size and structure of the U.S. economy, especially the manufacturing sector, but it
has also complicated the task of measuring economic growth and industry performance.
Policymakers and researchers are concerned that increased outsourcing to lower-cost offshore
suppliers has affected key economic measures such as output, value added, and labor input.
Difficulties in measuring price change for imported goods can affect measured growth in real
gross domestic product (GDP). Similarly, problems identifying outsourcing-related activities
could affect measures of industry contributions to economic growth and productivity in
manufacturing and other sectors.
For several decades, imports have been the major source of U.S. supply for final consumer goods
such as apparel, toys, shoes, motor vehicles, and consumer electronics, and for certain kinds of
business investment goods. More recently, the import share has increased for final goods such as
furniture and other household products, with important implications for the measurement of
domestic consumption prices. Another recent trend has been strong growth in the use of
imported intermediate materials by U.S. manufacturing industries, partly at the expense of
domestic goods. A significant portion of this trade occurs among affiliated parties within U.S.
multinational companies. This form of offshore outsourcing—substitution of imported for
domestic materials—has raised questions about the measurement of real value added by industry
and its impact on real GDP growth.
Limitations in the measurement of imports have somewhat different implications for the various
approaches typically used by statistical agencies to measure GDP.175 Only the final expenditures
approach and the production approach provide measures of both nominal and real GDP.
Because imports are subtracted using either approach, import growth has important measurement
implications. In the final expenditures approach, which is featured in the United States, real
GDP is an aggregate of personal consumption expenditures, private equipment and structures,
government consumption and investment, and exports less all imports, both final and
intermediate. In the production approach, which is the featured approach in many other
countries, real GDP is an aggregate of the real value added originating in all industries, including
government. Value added equals gross output less intermediate inputs, which include imported
inputs. As a result, under the production approach, only the imports consumed in intermediate
uses are subtracted. In recent years, intermediate goods and services have accounted for slightly
more than 50 percent of all imports.
175
The three major approaches are the final expenditures approach, the income approach, and the production
approach.
343
Economists have expressed concern that substitution by U.S. consumers and producers toward
lower-priced imported goods from developing countries may not be fully reflected in the official
import-price indexes used for calculating real GDP based on the final expenditures approach, and
that as a consequence, growth in real GDP and productivity may be overstated (Mandel). This
potential bias can be explored using both the expenditure and production approaches to
measuring real GDP. To the extent that the recent growth in lower-priced imported goods has
affected intermediate and final uses assessing the impact using the production approach may be
revealing.
Data from the Bureau of Economic Analysis’s (BEA) annual industry accounts can be used to
identify not only the uses of imported goods (intermediate vs. final) but also the overall
importance of imported products by measuring their value relative to the value of comparable
domestically produced goods. For this paper, we use data from the BEA’s annual industry
accounts and from the BEA’s surveys of multinational companies (MNCs) to determine how
growth in imported intermediate inputs has affected growth in real value added by industry (real
GDP growth), and to assess the impact of alternative assumptions about the use of imports and
the behavior of import prices.
In this paper, we calculate real value added by industry and real value added for all industries
(real GDP) using alternative assumptions about industry use of imports and the behavior of
imported input prices. In the current (baseline) methodology, the allocation of imports to
industries is based on an “import comparability” assumption. This assumes that the portion of
intermediate inputs attributable to imports is calculated as a percentage of the total purchase
value, using the economy-wide ratio of commodity imports to the total domestic supply of the
commodity. Alternative assumptions about the use of imports by selected industries are based
on the BEA’s data on imports by the U.S. parents of foreign affiliates and unaffiliated parties.
For the deflation of imported intermediate inputs, the current methodology relies primarily on
import-price indexes compiled by the Bureau of Labor Statistics (BLS). Alternative assumptions
about import-price change are made to determine a threshold required for import price biases to
impact real GDP, the manufacturing sector, and selected manufacturing industries.
The remainder of this paper is presented in four sections. The next section provides background
on how the industry accounts can be used to measure outsourcing and the role of imported inputs
at the industry level, how the BEA’s import-use tables are compiled, and how imported inputs
are used in constructing real value added by industry. After that we briefly describe the BEA’s
International Economic Accounts, including the MNC data, and explain how the MNC-based
import-use tables are compiled. The following section presents empirical results that compare
the current (baseline) estimates from the annual industry accounts with results from the MNCbased import-use tables. This section also describes results based on different assumptions about
the behavior of import prices. We conclude with a brief summary and recommendations for
improving data on imported inputs by industry and import-price indexes.
344
ANNUAL INDUSTRY ACCOUNTS
The annual industry accounts are a useful analytical framework for simulating the impact of
alternative assumptions about imports for several reasons. These accounts provide an annual
time series of nominal and real gross output, intermediate inputs, and value added for 65
industries defined according to the 1997 NAICS) (Moyer et al. 2004). They provide an
internally consistent set of industry production accounts that are integrated conceptually and
statistically with final expenditures and GDP from the national income and product accounts
(NIPAs).
The annual industry accounts are estimated within the framework of balanced make and use
tables, which allows for integrated analysis of industry output, inputs, employment, final
demand, and imports. The annual input-output (I-O) accounts provide a time series of detailed,
consistent information on the flows of goods and services that both comprise industry production
processes and that are included in final expenditures. Estimates of the supply of commodities are
prepared at nearly the same level of detail as in the benchmark I-O accounts and are then
aggregated to the higher publication level used for the annual industry accounts. The GDP by
industry accounts feature estimates of nominal and real value added by industry. Value added is
defined as an industry’s gross output (sales or receipts and other operating income) minus its
intermediate inputs (energy, materials, and purchased services). Intermediate inputs are acquired
from either domestic or foreign sources (imports). Price and quantity indexes of gross output,
intermediate inputs, and value added are published for industries, industry groups, and broad
sectors in the GDP by industry accounts.
Significant improvements in the measurement of intermediate purchases in the 1997 benchmark
made the annual industry accounts more suitable for identifying and measuring outsourcing and
the role of imported inputs than in the past.176 A broader set of purchased services was collected
for establishments in the manufacturing, mining, and construction sectors, and more detailed data
on purchased services for more industries in the trade and services sector were collected from an
expanded Business Expenses Survey.177 Estimates of materials and energy inputs by industry
were also based on detailed economic census data for manufacturing and on broader input
category data for nonmanufacturing industries.
The expansion of the annual industry accounts in 2005 to provide additional information on the
composition of intermediate inputs by industry made these accounts more analytically useful to
study trends in the use of energy, materials, and purchased services inputs (Strassner, Medeiros,
and Smith 2005). The balanced I-O use table provides the product detail needed to aggregate
estimates of intermediate inputs into cost categories useful for economic analysis.178 Each cost
category includes both imported and domestically produced goods and services, and each
176
The 1997 benchmark I-O accounts were based almost entirely on detailed data on outputs and inputs collected by
the Census Bureau in the 1997 Economic Census. For more information, see Lawson et al. (2002).
177
As a result of the expansion in source data, a much larger share of total intermediate purchased services was
based on economic census data than in past benchmarks.
178
These estimates are prepared by applying a KLEMS production framework to the BEA’s estimates of industry
production.
345
category is valued in purchasers’ prices, which include domestic transport costs, wholesale trade
margins, and sales and excise taxes.
On an annual basis, a wide array of source data is used to update the benchmark estimates for the
annual time series. Nominal value added by industry estimates are available annually for the
compensation of employees, taxes on production and imports less subsidies, and the gross
operating surplus. Annual survey data are available from the Census Bureau for updating
industry gross output for all of the manufacturing industries and for most of the services
industries, including the industries that provide outsourcing-related services. Annual data are
also available from the NIPAs for updating estimates of final expenditures.
Data, however, are not available annually to update estimates of intermediate inputs by industry.
Instead, the BEA’s procedures for annual updates of intermediate inputs rely partly on the
assumption that the real (constant-price) use of intermediate inputs relative to the industry’s real
gross output has not changed from the prior year.179 An industry’s real intermediate inputs are
initially updated based on changes in its real gross output. The nominal value of its intermediate
inputs for the current year is further adjusted based on price changes for the detailed commodity
inputs. Balancing constraints are imposed to ensure that the use of commodities by all industries
equals the supply of commodities, after accounting for final uses from the NIPAs.180 These
procedures are used for each year’s set of accounts. Updated KLEMS estimates by industry are
likewise based on the updated commodity input estimates.
An important step in updating the annual industry accounts is the development of import-use
tables to allow for the separate deflation of domestic and imported inputs in the calculation of
real value added.181 Intermediate inputs at a detailed product level are disaggregated to obtain the
domestic and imported portions of intermediate inputs included in each KLEMS input cost
category. For each detailed commodity used by an industry, the portion attributable to imports is
calculated as a percentage of the total purchase value, using the economy-wide ratio of
commodity imports to the total domestic supply of the commodity.182 Although this assumption
is necessary, the import content of specific types of goods and services could vary by industry as
a result of factors such as affiliation status, location, product mix, relative prices, or technology.
The BEA uses this approach because of the lack of actual data on the use of imports by
industry183
This distinction between domestic and imported inputs allows for differences in the behavior of
prices for imported and domestic products to be accounted for when separate price indexes are
179
This is often described as the “constant industry technology” assumption.
The annual I-O accounts use final expenditure categories from the NIPAs as controls during the biproportional
balancing of the use tables. An additional balancing constraint is that the sum of nominal value added across all
industries must equal GDP.
181
Fisher-ideal index number formulas are used to prepare chain-type indexes for gross output, intermediate inputs,
and value added by industry, and for higher-levels of aggregation. For more information, see the technical appendix
in Moyer et al. (2004).
182
For example, if imports represent 35 percent of the domestic supply of semiconductors, then the estimates in the
import-use table assume that imports comprise 35 percent of the value of semiconductors in each industry that uses
semiconductors.
183
This “import comparability assumption” is often used in studies of the impact of imports on intermediate inputs.
180
346
available.184 For domestic materials and for energy, the price indexes are mostly BLS producer
price indexes (PPIs), Department of Energy implicit price deflators, and price indexes from other
sources that are considered reliable. Many of the services input-price indexes are also obtained
from BLS PPIs, but some are based on other sources that are not as reliable, either because of
quality change or due to assumptions about labor productivity.185 Price indexes for imported
materials are largely based on the BLS International Price Index program. Price indexes for
imported services are much more limited in their coverage.
International Accounts
The BEA’s international transactions accounts (ITAs) provide monthly, quarterly, and annual
estimates of transactions between the United States and foreign residents.186 ITAs include a
current account, a capital account, and a financial account. The two major components of the
current account are 1) exports of goods and services and factor income receipts, and 2) imports
of goods and services and factor income payments. The difference between these two
components, plus net unilateral current transfers, equals the balance on the current account. The
capital account includes capital transfers such as debt forgiveness. The two major components
of the financial account are 1) changes in net U.S.-owned assets abroad, and 2) changes in net
foreign-owned assets in the United States. These components are the major source of change in
the United States’net international investment position.
The BEA also produces comprehensive statistics on U.S. direct investment abroad and foreign
direct investment in the United States that are required for compiling ITAs and for analysis of
MNCs. The BEA’s data on MNCs are potentially very useful for assessing assumptions about
the use of imported goods by industries because these companies account for about 60 percent of
U.S. imports of total intermediate inputs. While imports of goods in ITAs are based primarily on
data compiled by the Census Bureau from import shipping documents, the data on imports of
goods reported to BEA on the MNC surveys conform well to Census Bureau concepts and
definitions. Imports of services in ITAs are estimated from a variety of sources, primarily the
BEA’s surveys of U.S. and foreign MNCs and the BEA’s surveys of U.S. international services
transactions between unaffiliated parties. For this study, we used annual data on U.S. imports of
goods and services shipped to the U.S. parents by both their foreign affiliates and other
unaffiliated parties.
The BEA’s surveys are mandatory and collect selected data for transactions between the U.S.
parents of MNCs and both their foreign affiliates and unaffiliated parties and transactions
between the U.S. affiliates of foreign MNCs and both their foreign parent companies and certain
other affiliated foreign firms. These data play an important role in compiling ITAs and in
providing additional detail on cross-border trade in services and on services provided by the
affiliates of MNCs. Because U.S. MNCs are typically very large firms, the combined data for
U.S. parents and U.S. affiliates of foreign MNCs account for a significant share of domestic
184
Domestic prices are used to deflate imported inputs in cases where import prices are unavailable.
Expansion of the BLS PPI program in the services sector during the 1990s has resulted in better coverage and
improved quality, but gaps and limitations remain.
186
Transactions between the United States and its territories, Puerto Rico, and the Northern Mariana Islands are not
treated as foreign transactions in ITAs.
185
347
economic activity, especially in the goods-producing sector of the economy. These combined
company data, when classified by industry, provide valuable insights into the industry
distribution of imports.
For this paper, alternative import-use tables were constructed using data from the BEA’s surveys
of MNCs. 187 Companies in these surveys are classified according to the International Survey
Industry (ISI) classification system, a system developed by the BEA that is based on the NAICS.
The MNC surveys provide information on total imports by U.S. parent firms (from both
affiliated and unaffiliated parties) classified by the U.S. parents industry, and imports by U.S.
parent firms from foreign affiliates classified by the foreign affiliate’s industry. For the
“benchmark” survey years, additional product information on imports is provided. These broad
product categories are listed below:
•
Food, live animals, beverages, and tobacco
•
Crude materials, inedible, except fuels
•
Mineral fuels, lubricants, and related materials
•
Chemicals and related products
•
Industrial machinery and equipment
•
Office machines and automatic data processing machines
•
Telecommunications, sound equipment, and other electrical machinery and parts
•
Road vehicles and parts
•
Other transportation equipment
•
Other products
Linking the foreign affiliate to its U.S. parent provides the basis for a commodity and industry
classification for the import-use framework. Industry classification is based on the ISI industry
of the U.S. parent and commodity classification is based on the ISI industry of the foreign
affiliate. To develop this mapping, imported products from the foreign affiliate were compared
to the ISI industry of the foreign affiliate. In most cases, the ISI industry of the foreign affiliate
aligns well with the product imported. For example, a foreign affiliate classified in
pharmaceuticals and medicines manufacturing (ISI 3254) ships products categorized in
chemicals and related products (NAICS 325). Because industry classifications are not available
for unaffiliated parties, imports from unaffiliated firms are assumed to resemble those of
affiliated firms. The import-use tables based on ISI industry categories were converted to the
1997 NAICS-based structure used for the Annual Industry Accounts. Import shares for
commodities purchased by each industry were calculated as the ratio of the commodity import
value for that industry and the total import value for that industry. In total, the MNC-based
187
The estimates are based on special tabulations prepared by the BEA’s Direct Investment Division (DID). The
DID provided access to databases that allowed the authors to identify and tabulate imported goods and services
directly.
348
imports accounted for about 60–65 percent of all imported intermediate inputs presented in the
annual industry accounts.
EMPIRICAL RESULTS
The existing framework and methodology for the annual industry accounts was used to prepare a
set of baseline estimates that can be compared with the results of simulation exercises that
incorporate alternative assumptions about import use and prices. These results focus on
industries that are the largest users of imported goods, such as computers and electronic products
and chemicals manufacturing. Assumptions about both the use of imports and the behavior of
import prices are important because an industry’s real value added is calculated as the difference
between real gross output and real intermediate inputs.188
Table 1 shows that, in the aggregate, the import comparability assumption provides results that
are largely consistent with actual data on the use of imports by industry from the BEA’s MNC
surveys. These results indicate that the assumptions underlying the industry distributions of
imported inputs in the annual industry accounts give reasonable results at aggregate levels, but
that improvements are possible at more detailed industry levels. Some differences in the results
at detailed levels are attributable to the fact that the data from the International Accounts are
classified by industry on an enterprise basis, whereas data from the annual industry accounts are
classified by industry on an establishment basis. Within both the goods- and services-producing
sectors, some large share differences for industry groups are offset at higher levels of
aggregation, suggesting the possibility that the differences are attributable largely to differences
in classification.
188
Estimates of real value added by industry are affected by both the source of the inputs and the import price
indexes used for deflation. For example, if the computer manufacturing industry uses more imported
semiconductors than assumed and if import prices are falling faster than domestic prices, or if the actual price of
imported semiconductors is falling faster than the official import price index, then real intermediate input is
understated and real value added is overstated in the computer manufacturing industry.
349
Table 1. Import Shares by Industry, 2002
(Comparison of International Accounts and Annual Industry Accounts Import-Use Tables)
(Percent)
International
Accounts
Annual Industry
Accounts
Manufacturing
Distributive services/1/
Information
Finance, insurance, real estate, rental, and leasing
Professional and business services
Other industries/2/
20.7
3.3
4.2
0.9
2.3
6.5
16.8
7.0
5.3
5.0
3.9
6.4
Addenda
Private goods-producing industries/3/
Private services-producing industries/4/
17.7
3.5
14.9
5.4
Industry Group
/1/ Consists of w holesale trade; retail trade; transportation and w arehousing
/2/ Consists of agriculture, forestry, fishing, and hunting; mining; construction; educational services; health
care and social assistance; arts, entertainment, and recreation; accommodation and food services;
and other services, except government
/3/ Consists of agriculture, forestry, fishing, and hunting; mining; construction; and manufacturing.
/4/ Consists of utilities; w holesale trade; retail trade; transportation and w arehousing; information; finance
and insurance; real estate and rental and leasing; professional, scientific and technical services;
management of companies and enterprises; administrative and w aste management services;
educational services; health care and social assistance; arts, entertainment, and recreation;
accommodation and food services; and other services, except government.
Over the period 1999–2006, import shares of materials for manufacturing based on the MNC
data are consistently higher than those constructed for the industry accounts using the import
comparability assumption; however, the pattern of growth between the two series is similar
(Chart 1). On average, the MNC data suggest that the annual industry accounts understate
import shares of materials inputs by about 4 percentage points per year for manufacturing.
Chart 1. Imported Materials Inputs as a Share of Total Intermediate
Inputs for Manufacturing, 1999-2006
25.0
Percent
20.0
15.0
10.0
5.0
0.0
1999
2000
2001
2002
2003
Annual Industry Accounts
2004
2005
2006
Within manufacturing, the composition of imported inputs for materials shows some variation.
Table 2 presents import shares for manufacturing at the commodity level constructed for the
350
annual industry accounts compared with those constructed using MNC data. There are notable
differences in import shares across the board. For 10 of the largest annual industry accounts’
publication-level commodities within manufacturing, the largest differences are shown for oil
and gas extraction and computers and electronic products. Most other commodities show much
smaller differences.
Table 2. Imported Input Shares for Manufacturing Commodities, 2002
(Percent)
Commodity
Oil and gas extraction
Computer and electronic product manufacturing
Primary metal manufacturing
Chemical manufacturing
Machinery manufacturing
Fabricated metal product manufacturing
Paper manufacturing
Electrical equipment, appliance, and component manufacturing
Food manufacturing
Wood product manufacturing
International
Accounts
36.9
70.6
8.5
26.4
26.2
3.8
2.6
24.7
9.9
16.2
Annual Industry
Accounts
63.6
42.9
20.7
13.4
29.0
10.2
13.8
35.2
12.2
24.0
One would expect that changing the mix of intermediate inputs sourced from domestic versus
foreign production could lead to important differences in price growth for imported intermediate
inputs. Charts 2 and 3 show chain-type price indexes for economy-wide energy, materials, and
purchased services inputs based on import shares developed for the annual industry accounts and
those based on the MNC data. Price growth for imported energy inputs and purchased services
inputs increases at a slower pace using data from the annual industry accounts; price growth for
materials inputs increases at a faster rate.
Chart 2. Chain-Type Price Indexes for Imported Intermediate Inputs, 1999-2006
(Annual Industry Accounts Import-Use Table)
Indexes (2000 = 100)
200
160
120
80
40
1999
2000
2001
Energy
2002
2003
Materials
351
2004
Services
2005
2006
Chart 3. Chain-Type Price Indexes for Imported Intermediate Inputs, 1999-2006
(International Accounts Import-Use Table)
Indexes (2000 = 100)
200
160
120
80
40
1999
2000
2001
Energy
2002
2003
Materials
2004
2005
2006
Services
Within manufacturing, the trend for materials inputs is more focused. Over 1999–2006, price
growth for imported materials inputs for manufacturing increases at a faster pace using data
constructed for the annual industry accounts than that based on the MNC data (Chart 4). The
slower materials price growth resulting from higher overall import shares for materials
constructed with MNC data, coupled with BEA’s existing import prices, suggest real
intermediate inputs in the annual industry accounts may be understated and, therefore, that real
value added for manufacturing is overstated.189
Chart 4. Chain-Type Price Indexes for Im ported Materials for Manufacturing
(Com parison of International Accounts to Annual Industry Accounts Im port-Use Table)
Indexes (2000 = 100)
160
140
120
100
80
60
1999
2000
2001
2002
2003
2004
2005
2006
However, because existing import price data is incomplete, changing the sourcing mix for
intermediate inputs does not impact the price indexes for intermediate inputs for all industries or,
even, the manufacturing sector (Charts 5 and 6). Some differences do exist for high-import
189
In addition to materials inputs, real intermediate inputs growth is a function of the source and price mix for
energy and purchased services inputs. Therefore, compositional effects within intermediate inputs would have to be
examined to determine how real intermediate inputs have changed. Nevertheless, manufacturing is a high importer
of materials inputs relative to imports of energy and purchased services inputs.
352
industries, such as computer and electronic products, where existing good-quality import-price
data is used to deflate the import content of intermediate inputs.
Chart 5. Chain-Type Price Indexes for All Indus trie s' Total Interm ediate Inputs
(Com parison of International Accounts to
Annual Industry Accounts Im port-Use Table)
Indexes (2000 = 100)
130
120
110
100
90
80
1999
2000
2001
2002
2003
2004
2005
2006
Chart 6. Chain-Type Price Indexes for Total Interm ediate Inputs for Manufacturing
(Com parison of International Accounts to
Annual Industry Accounts Im port-Use Table)
Indexes (2000 = 100)
130
120
110
100
90
80
1999
2000
2001
2002
2003
2004
2005
2006
The limited availability of import-price data, overall, is the primary reason that changing the mix
of import shares does not impact the aggregate growth rates. There are 1,028 item-level building
blocks used to deflate imported intermediate inputs in the annual industry accounts, but only 57
percent have import prices available for use in the separate deflation of intermediate inputs. The
remaining item-level goods and services are deflated with domestic price indexes. The mix of
import-price coverage also differs by sector, with about 58 percent coverage for the Goods sector
and 50 percent for the Services sector. Finally, import-price coverage does not necessarily imply
a good match, as many of the item-level building blocks covered are deflated with an aggregate
import-price index.
Given data limitations for import-price indexes available from the U.S. statistical system, we
conducted several simulations to determine the threshold of import price biases that are required
to affect real value added growth across all industries (real GDP), manufacturing, and a selected
number of manufacturing industries. As a first step, we adjusted the domestic prices used to
deflate the import content of intermediate inputs by applying the average price differential that
353
exists for item-level detail when both domestic and import-price indexes are available. This
difference averaged about 0.4 percent per year.
This bias adjustment translates into notable differences in price growth for imported intermediate
inputs for All industries and manufacturing. Over the period 1999–2006, price growth for
economy-wide imported intermediate inputs increases at an average annual rate that is about 1.5
percentage points slower when import shares are based on MNC data instead of the annual
industry accounts. Price growth is about 3 percentage points slower for manufacturing (Charts 7
and 8). The impact, however, on real GDP growth and real value added for manufacturing is
negligible: real GDP grew at an average annual rate of 2.5 percent using MNC-based import
data, compared to 2.6 percent for the annual industry accounts, and real value added growth for
manufacturing grew 1.9 percent and 2.0 percent, respectively.
Chart 7. Chain-Type Price Indexes for All Industries' Imported Intermediate Inputs
with 0.4 Percent Import Price Bias Adjustment
Indexes (2000 = 100)
140
120
100
80
1999
2000
2001
2002
2003
2004
2005
2006
Chart 8. Chain-Type Price Indexes for Manufacturing's Imported Intermediate
Inputs with 0.4 Percent Import Price Bias Adjustment
Indexes (2000 = 100)
140
120
100
80
1999
2000
2001
2002
2003
2004
2005
2006
Finally, we applied a series of bias adjustments, ranging from a 1 percent to 5 percent
overstatement of import prices, to determine the thresholds required to affect real GDP growth,
and real value added growth for manufacturing, computer and electronic products, chemical
manufacturing, and machinery manufacturing. The 1–5 percent bias adjustments were applied to
354
all existing import prices and to the domestic prices that are used to deflate the import content of
imported intermediate inputs for goods and services that do not have import price coverage.190
The results in Table 3 show average annual growth rates for real value added based on the
various bias adjustments applied to the existing annual industry accounts import-use tables and
those constructed using MNC data.
Overall, each 1percent bias adjustment to import prices led to an average annual decrease of 0.08
percentage points for all industries, 0.33 for manufacturing, 0.46 for computer and electronic
products, 0.22 for chemicals, and 0.24 for machinery manufacturing, when using import shares
based on the annual industry accounts. One percent bias adjustments, when using import shares
based on MNC data, led to an average annual decrease of 0.08 percentage points for all
industries, 0.39 for manufacturing, 0.60 for computer and electronic products, 0.27 for
chemicals, and 0.18 for machinery manufacturing.
Table 3. Average Annual Growth Rate for Real Value Added by Industry with Price Adjustments, 1999 - 2006
(Percent)
Industry
Baseline 0.04%
All Industries
Manufacturing
Computer and electronic products manufacturing
Machinery manufacturing
Chemical manufacturing
Oil and gas extraction
1%
3%
5%
2.6
2.6
2.6
2.5
2.5
2.5
2.4
2.3
2.2
2.2
2.2
2.0
2.0
1.9
1.8
1.6
1.2
0.9
0.5
0.1
17.2
17.2
17.1
17.1
16.8
16.8
15.8
15.6
14.9
14.4
1.4
1.4
1.3
1.3
1.2
1.2
0.7
0.9
0.2
0.5
2.6
2.5
2.6
2.4
2.4
2.2
2.0
1.7
1.5
1.1
-3.4
-3.2
-3.5
-3.2
-3.5
-3.3
-3.7
-3.4
-3.8
-3.6
SUMMARY AND CONCLUSION
In the annual industry accounts, imports of intermediate inputs are constructed using the import
comparability assumption for purposes of separately deflating domestic and imported
intermediate inputs in the calculation of real value added. An analysis of import shares for the
190
Each bias adjustment decreased import prices used in the deflation by 1–5 percent.
355
annual industry accounts compared to import shares constructed using actual source data from
the International Accounts shows that the import comparability assumption provides a good
approximation of imported intermediate use. Differences at detailed levels, however, may be the
result of using MNC company data rather than establishment-based data. In addition, while
MNC imports account for about 60–65 percent of total imported intermediate inputs in the
annual industry accounts, it is possible that import usage differs among the smaller firms that
account for the remaining 35–40 percent. The possible impact of this coverage difference may
be worth examining more closely.
Most notably, data from the international accounts suggest that growth in real imported materials
inputs is likely understated in the annual industry accounts. However, this understatement does
not currently lead to large differences in real value added growth for all industries (real GDP) or
for manufacturing because of the limited availability of import-price data used to deflate the
import-content of intermediate inputs. A simulation of a range of bias adjustments for import
prices used in deflation suggests that better import-price measurement will improve the accuracy
of real value added by industry; however, the overall magnitude of the bias adjustments would
need to be about 6.5 percent to affect real GDP growth by at least one-half of a percentage point,
irrespective of whether import shares are from the annual industry accounts or based on data
from the international accounts.
Further study is required to develop a better understanding of how imported inputs affect
industry output, employment, real value added, and contributions to GDP. More research is also
needed to determine the sensitivity of these results to the assumptions used by the BEA for the
annual industry accounts with respect to the classification of imported goods and services, the
distribution of goods and services by using industry, and the behavior of import prices. The
BEA will continue to review these assumptions and will further investigate company-based data
from the international accounts that could help evaluate the assumptions underlying the industry
distributions. The BEA is also interested in working with the BLS International Price Program
to try to develop improved price indexes for the deflation of imported intermediate inputs in both
the NIPAs and annual industry accounts. The BEA is also interested in the idea of input-price
indexes that are proposed by the BLS. These input-price indexes could be used to deflate total
intermediate inputs without concern for the sourcing mix. Input-price indexes, in conjunction
with domestic price indexes, could be used to calculate import price indexes, allowing for the
continued study of imported intermediate inputs.
The BEA plans to investigate these differences in more detail with the goal of obtaining
improved industry distributions of imported intermediate inputs in the annual industry accounts.
Better grounding of these assumptions is important not only for understanding the role of
imported inputs in the U.S. economy, but also for developing more reliable quantity and price
indexes for intermediate inputs and value added by industry.
356
References
Lawson, Ann M., Kurt S. Bersani, Mahnaz Fahim-Nader, and Jiemin Guo. 2002. “Benchmark
Input-Output Accounts of the United States, 1997.” Survey of Current Business 82
(December):19–109.
Mandel, Michael. 2007. “The Real Cost of Offshoring.” Business Week, June 18.
Moyer, Brian C., Mark A. Planting, Mahnaz Fahim-Nader, and Sherlene K.S. Lum. 2004.
“Preview of the Comprehensive Revision of the Annual Industry Accounts: Integrating the
Annual Input-Output Accounts and the Gross-Domestic-Product-by-Industry Accounts.”
Survey of Current Business 84(March): 38–51.
Strassner, Erich H., Gabriel W. Medeiros, and George M. Smith. 2005. “Annual Industry
Accounts: Introducing KLEMS Input Estimates for 1997–2003.” Survey of Current Business
85 (September): 31–65.
Yuskavage, Robert E., Erich H. Strassner, and Gabriel W. Medeiros, 2008. “Domestic
Outsourcing and Imported Inputs in the U.S. Economy: Insights from Integrated Economic
Accounts.” 2008. Paper prepared for the 2008 World Congress on National Accounts and the
Economic Performance Measures for Nations.
———. 2009. “Outsourcing and Imported Services in BEA’s Industry Accounts.” In
International Trade in Services and Intangibles in the Era of Globalization, Marshall
Reinsdorf and Matthew J. Slaughter, eds. Chicago: University of Chicago Press.
357
Evaluating Estimates of Materials Offshoring
from U.S. Manufacturing
*
by
Robert C. Feenstra
University of California, Davis and NBER
J. Bradford Jensen
Georgetown University and NBER
October 2009
*
The authors thank Greg Wright, UC Davis, who prepared portions of this paper and Daniel Reyes, Georgetown,
for research assistance. We thank Belinda Bonds of the Goods and Distributive Services Branch at the Bureau of
Economic Analysis for assistance. Some of the analysis in this paper was conducted at the Center for Economic
Studies, U.S. Census Bureau. The results have not undergone the review accorded Census Bureau publications and
no endorsement should be inferred. Any opinions and conclusions expressed herein are those of the author(s) and do
not necessarily represent the views of the U.S. Census Bureau or the National Bureau of Economic Research. All
results have been reviewed to ensure that no confidential information is disclosed.
359
When materials offshoring is measured by estimating imported intermediate inputs, a common
assumption used is that an industry’s imports of each input, relative to its total demand, is the
same as the economy-wide imports relative to total demand: this is the so-called “import
comparability assumption” (Houseman 2008, p. 9), or the “proportionality assumption ” (OECD
STAN database). That assumption was made by Feenstra and Hanson (1999), for example, and
was critiqued by the National Research Council ([NRC] 2006) as being a significant limitation of
current data collection and analysis. Recent work by Winkler and Milberg (2009) for Germany
shows that this assumption does not hold up well when compared to the actual imports by
industries. For the United States, too, it is highly desirable to move beyond this assumption to
obtain a direct measure of imported materials by industry.
The goal of this project is to obtain such an industry-level measure of offshoring for the United
States. We begin, however, with a smaller first step. In the first step, we explore alternatives to
the Feenstra-Hanson (1999) measure of offshoring that still make use of the import
comparability assumption. While that measure of offshoring was intended to reflect imported
intermediate inputs, in practice it also included imported final goods. So in this first step, we
recalculate the Feenstra-Hanson (1999) measure of offshoring while focusing on only imported
intermediate inputs as defined by end-use classifications. This approach has been taken by
several other recent authors, including Bergstrand and Egger (2008), Sitchinava (2007, 2008),
and Wright (2009).
In the second step, we explore a different methodology for allocating imported inputs across
industries using firm-level data on imports and production. We use information on imports from
the Linked/Longitudinal Firm Trade Transaction Database (LFTTD), which links individual U.S.
trade transactions to firms and information on materials used and products produced from the
census of manufactures. We use the linked production and import data to construct firm-level
input-output (I-O) tables and then aggregate these to the industry level to derive imported input
intensity by industry and compare our results with those obtained by the BEA using the “import
comparability” assumption. Our focus is on imports of intermediate inputs, so we again use the
end-use classification to exclude from the analysis products identified at “final goods” by Wright
(2009). We confront a number of technical and data issues and make several compromises as a
result, all of which we describe in the paper. We describe differences between the import matrix
constructed using firm-level import data and BEA’s import matrix.
THE IMPORT COMPARABILITY ASSUMPTION
Our goal is to update the offshoring measure described in Feenstra and Hanson (1996, 1999),
which is defined for any industry k purchasing inputs j as
Industry k share of intermediate inputs that are imported
⎛
(1)
=
imports of good j
⎞
∑ j (industry k purchases of good j ) ⎜ total domestic consumption of j ⎟
⎝
∑ j (industry k purchases of good j )
361
⎠
The primary shortcoming of this measure is the use of good j’s share of imports in total domestic
consumption, in the numerator, which is computed for the entire U.S. economy. Obviously, it
would be preferable to measure this share by just using data for purchasing industry k, as we
shall attempt to do in the second step of this project. As it is stated, Equation (1) essentially
assumes that the economy-wide import share for good j is the same as the industry k import share
for good j, which is the “import comparability” assumption.
Given this limitation of Equation (1), there are still some improvements that can be considered
before using firm-level data. Specifically, we consider recalculating the measure of offshoring in
Equation (1) while focusing more carefully on only imported intermediate inputs. Specifically,
the inputs j that are used in Equation (1) are defined by the classifications used in I-O tables of
the United States: either 4-digit SIC before 1996 or 6-digit North American Industry
Classification System (NAICS) after 1996. For each of classifications, there will be multiple 10digit Harmonized System (HS) imported products. Let us denote by i ∈ I j the set of 10-digit HS
products within each 4-digit SIC before 1996 or 6-digit NAICS good i. Then a more accurate
definition of the Feenstra and Hanson (1999) measure of materials offshoring is:
Industry k share of intermediate inputs that are imported
⎛
(1')
=
sum over imports i ∈ I j
∑ j (industry k purchases of good j ) ⎜⎜ total domestic consumption i ∈ I
⎝
(industry
k
purchases
of good j )
∑j
⎞
⎟⎟
j⎠
A problem with this definition of offshoring is that some of the imported products i can be final
goods rather than intermediate inputs. Imports of such final goods are often not what we have in
mind with materials offshoring. To correct this problem we can restrict attention to HS goods
with corresponding “end-use codes” that are indeed intermediate inputs. The end-use codes are
used by the Bureau of Economic Analysis (BEA) to allocate goods to their final use, within the
National Income and Product Accounts. Accordingly, U.S. imports and exports by Harmonized
System are also allocated to end-use codes. As described by the Census Bureau, Guide to
Foreign Trade Statistics:191
The 1-digit level end-use categories provide data for the following broad aggregates: (0)
Foods, feeds, and beverages; (1) Industrial supplies and materials; (2) Capital goods,
except automotives; (3) Automotive vehicles, parts and engines; (4) Consumer goods
(nonfood), except auto; and (5) Other merchandise.
…The HTSUSA and Schedule B classifications are summarized into six principal "enduse" categories and further subdivided into about 140 broad commodity groupings. These
categories are used in developing seasonally adjusted and constant dollar totals. The
concept of end-use demand was developed for balance of payments purposes by the
Bureau of Economic Analysis.
191
Slightly amended from http://www.census.gov/foreign-trade/guide/sec2.html.
362
Based on the numbering system defined in the above quotation, food and other items begin with
the digit 0, which include both final goods and intermediate inputs; raw materials and
intermediate goods begin with 1; investment goods begin with the digit 2; automotive goods
begin with 3, which include both final goods (finished autos) and intermediate inputs (parts);
final consumer goods (nonfood) begin with the digit 4; and 5 is a miscellaneous category. In
Appendix A we list the precise 5-digit end-use codes that are included within final goods (i.e.,
consumption and investment), while all other end-use codes are treated here as intermediate
inputs or raw materials.192
Using this end-use classification, we consider a restricted set of HS codes within each SIC or
NAICS industry j
I j ≡ {HS goods i within the industry j that are also intermediate inputs} .
Then the revised measure of materials offshoring is
⎛
(2)
sum over imports i ∈ I j
∑ j (industry k purchases of good j ) ⎜⎜ total domestic consumption i ∈ I
⎝
∑ j (industry k purchases of good j )
⎞
⎟⎟
j⎠
.
Note that the import share used in the numerator of Equation (2) restricts the set of goods used in
both the numerator and the denominator, so we cannot tell how it compares with the import share
used in Equation (1'). Specifically, the denominator of this import share is constructed as
total domestic consumption i ∈ I j
= domestic shipments for i ∈ I j + sum over imports i ∈ I j – sum over exports i ∈ I j .
The import and export terms in this expression do not need any explanation: they are simply the
sum over HS imports or exports within the SIC or NAICS industry j, that are also intermediate
inputs (as defined by their end-use classification). But the domestic shipments term does require
an explanation. Rather than use the total domestic shipments of industry j, we instead
apportioned those domestic shipments into various HS products i, by assuming that the share of
domestic shipments for each HS product i within industry j equals the share of U.S. exports in
that HS product and industry. We then sum domestic shipments over just those HS products that
are also intermediate inputs (as defined by their end-use classification).
Empirical Implementation
We construct the offshoring measure Equation (2) for all years between 1980 and 2006 within
the manufacturing sector. We begin with measures of intermediates purchases by U.S. industries,
192
We thank Marshall Reinsdorf, BEA, for providing the end-use classifications in Appendix A. As noted in the
appendix, certain raw materials such as oil and minerals are always excluded from the offshoring calculation.
363
which are obtained from the economic census for benchmark years (1982, 1987, 1992, 1997,
2002). Prior to 1997, values are by 4-digit SIC codes and post-1996 values are by 6-digit
NAICS. Each observation in the economic census benchmark dataset contains a purchasing
industry, a corresponding intermediate industry which provides inputs, and a total value of
purchases (inputs). To obtain purchases for all years for an industry from a particular
intermediate industry, we simply interpolate and extrapolate the benchmark values linearly
throughout the period 1980—2006.193
The next step is to construct the import share of intermediates in domestic consumption of
intermediates. This industry share will be merged with the input-providing industries from the
purchases data described above. First, we merge data on imports and exports from Feenstra
(1996) and Feenstra, Romalis, and Schott (2002) with yearly data on total industry shipments,
obtained from the Annual Survey of Manufactures. Again, prior to 1997 these data are by 4-digit
SIC and post-1996 by 6-digit NAICS, so the merge is straightforward.
Now, in order to restrict the imports, exports, and shipments to intermediates only, we use the
end-use categories that are matched to SIC and NAICS industries in the import/export datasets.
The end-use categories that we excluded because they are “final goods” come from a list
provided by the BEA, as shown in the appendix. We separate investment goods and most
automobile categories from the list because these include many things that we think of as
vulnerable to offshoring, such as automobile parts and machinery and equipment, and therefore
we ultimately would like to include these items. For personal consumption expenditure (PCE)
goods, a portion of the list is more subjective, with some categories split between intermediate
and final goods. Here we simply remove all end-use categories that encompass some final
goods, and since the categories that are problematic are primarily food items, which we don’t
generally associate with offshoring activities, this approach seems reasonable. In addition, we
remove certain raw materials detailed in the appendix, such as petroleum products and various
metals, whose value and import volumes are likely unrelated to offshoring activities.
Table 1 shows trends in the offshoring measure using the original method of Feenstra and
Hanson (1990, equations [1] or [1']). We report both a broad and a narrow offshoring measure as
in Feenstra and Hanson (1996, 1999), where the narrow measure restricts the final and
intermediate industries to be within the same 3-digit NAICS categories. In comparison, Table 2
details trends in the revised offshoring measure, Equation (2), with and without inclusion of
investment goods.
We have compared the original and revised offshoring measures to determine which industries
show the greatest differences (averaged over years) and to obtain the results:
NAICS 339931: Dolls and stuffed toys, difference ≈ 0.85
NAICS 315991: Hats and caps, difference ≈ 0.35
NAICS 331316: Aluminum extruded products, difference ≈ 0.35
NAICS 311320: Chocolate and confectionary products, difference ≈ 0.29
193
The 2007 benchmark will be available beginning in June, and we will be able to reduce error caused by the
extrapolation of the 2002 benchmark.
364
NAICS 339941: Pens and mechanical pencils, difference ≈ 0.28
NAICS 339992: Musical instruments, difference ≈ 0.25
The industries with the greatest difference are simply consumer items that are imported directly
to retail outlets, so these imports are clearly final goods and therefore are omitted from the
revised offshoring measure.
ASSIGNING IMPORTED INPUTS TO INDUSTRIES USING FIRM-LEVEL DATA
In this section, we explore an alternative methodology to the “import comparability assumption”
for allocating imported inputs across industries. This alternative methodology uses transaction
data on firms’ imports linked to production data at the firm and plant level to construct
something analogous to firm-level I-O tables and then aggregates the firm-level I-O tables to
produce an aggregate import matrix that allocates imported intermediate inputs (I-O
commodities) across industries. This approach offers promise in that it provides a different
perspective on the allocation of imported intermediate inputs across industries. Our objective in
the remainder of the paper is to explain the alternative methodology, describe some of the
challenges we faced in trying to produce this, and then attempt to characterize (within the limits
of disclosure) how our alternative import matrix differs from the import matrix provided by the
BEA.
Data Used and Assignment Methodology
We use information on imports from 1997 from the Linked/Longitudinal Firm Trade Transaction
Database (LFTTD), which links individual U.S. trade transactions to firms (see Bernard, Jensen,
and Schott [2009] for more details on the LFTTD). We use information on materials used and
products produced from the 1997 census of manufactures. Because both datasets contain a firmlevel identifier, it is possible to link imported inputs to production data (materials used and
products produced) of the firms that import the intermediates. We will use this information to
construct a firm-level I-O table that allocates imported inputs across the products (I-O industries)
the firm produces.
The first limitation we confront is that the LFTTD contains firm-level identifiers for 80–85
percent of import value (roughly 10 percent of value is associated with transactions that have no
Employer Identification Number). As a result, our estimates of total import value across
commodities are systematically lower than the BEA’s, and this difference varies across
commodities. When we compare our allocation of imports across industries to BEA’s import
matrix, we will compare the shares of imports by industry (instead of levels) between the two
methodologies to try to mitigate the impact of this problem.
The LFTTD contains information on products at the 10-digit Harmonized System (HS)
classification level. We use publicly available concordances between 10-digit HS products and
6-digit BEA I-O commodities and assign 6-digit I-O commodity codes to all firm-level imports.
We then have firm imports on an I-O commodity basis.
365
The next step is to allocate the firm’s imports to the industries that use the imports in production.
Our intention was to use information collected in the census of manufactures regarding materials
used by manufacturing establishments as a way of allocating the use of imported intermediate
inputs to industries. The information on materials used is contained in “material trailers” in the
census of manufactures files and is classified using internal Census Bureau material codes. To
use this information to allocate imported commodities, we needed a bridge between internal
Census Bureau materials codes and I-O commodity codes. We obtained internal BEA
concordances between census material codes and BEA I-O codes.194 Using these codes, we were
able to allocate materials used to I-O industries.195 The assignment of materials used to industries
identified another limitation of our methodology—the value reported in the material trailers
accounts for only about 75 percent of the total materials used in the BEA I-O
tables.196
Another complication we confronted in allocating imported inputs to industries is that the vast
majority of trade value is mediated by large, multiunit, multiactivity firms (see Bernard, Jensen,
and Schott [2009]). Many of these firms have establishments classified in a range of sectors, e.g.,
the manufacturing sector, the wholesale sector, and the retail sector. Allocating imports across
industries within these firms proved difficult. One source of the difficulty is that the materialsused information is not collected in the same way for sectors outside of manufacturing, so we
needed some way to allocate commodity imports across industries. We tried allocating based on
the share of a firm’s total sales each establishment accounted for but found we were allocating
significant value for end-use commodities to manufacturing establishments owned by multisector
firms. To mitigate this problem, we excluded from our analysis 10-digit HS products classified
as “end use” by Wright (2009) and restricted our analysis to the manufacturing establishments of
importing firms.
After excluding end-use products and nonmanufacturing establishments/firms, we were able to
allocate 50 percent of total imported intermediate input value to manufacturing establishments
using the material codes (i.e., imported intermediate inputs were assigned to an establishment’s
I-O industry if the establishment reported using the material); the remainder of imported
intermediate value was allocated based on establishments’ share of a firm’s manufacturing
output. Because we exclude establishments within a firm that are outside of manufacturing, it is
possible that we overallocate imports to firms that have both manufacturing operations and
import for wholesale or retail operations. This highlights another potential compromise in our
methodology—that a significant share of imported intermediate value is imported by firms
whose manufacturing establishments do not report using the material.197
194
We thank Belinda Bonds of the BEA for providing the internal version of the concordance.
We used a concordance between NAICS industry classifications and BEA I-O industry classifications.
196
The BEA uses other sources and methodologies for constructing the materials used in the I-O tables.
Conversations with BEA staff suggested that our finding that the material trailers accounted for 75 percent of the IO value was in the right ballpark. There is a high correlation across commodity-industry cells between our materials
used values and the BEA I-O tables materials used.
197
It is difficult to know how to interpret this fact. One possibility is that establishments under-report the materials
that they use. An alternative explanation is that firms import a significant share of intermediate inputs that they do
not use for production.
195
366
With the assignment of imported intermediate inputs (I-O commodities) to establishments
that are classified by industry (I-O industry), we have essentially created firm-level I-O tables
with I-O commodity by I-O industry cells. The final step is to aggregate these firm-level cells to
obtain an import matrix for the manufacturing sector.
Comparison of BEA’s Import Matrix to Our Alternative Import Matrix
In this section, we attempt to characterize whether the allocation of imported intermediate inputs
differs between BEA’s import matrix, which uses the import comparability assumption, and our
alternative matrix, which uses firm-level data to assign commodities to industries—and, if so,
where it differs.198 Assessing whether the matrices differ in a meaningful way is obviously a bit
subjective and would depend to some extent on the purpose for which the matrix would be used.
We present descriptive statistics that attempt to quantify and characterize the differences between
the matrices from the two methodologies.
We focus on the share of a commodity’s import value assigned to a particular industry instead of
the level of import value to mitigate the issue posed by the systematic underallocation of
imported inputs in our data. To make the comparison, we exclude from BEA’s import matrix
industries that are outside of manufacturing and exclude products classified as end use (the same
products we excluded from the import data).199 For the manufacturing sector, we calculate the
share of an I-O commodity’s total imports that is allocated to each I-O industry within
manufacturing. We compare the shares in these I-O-commodity I-O-industry cells.
We begin by examining the simple correlation between the share of each 3-digit commodity
group’s total import value assigned to a 3-digit I-O-industry cell in the two matrices. The simple
correlation and BEA-value-weighted correlation are reported in Table 3. There is a high
correlation between the share in both the simple correlation and the weighted correlation; the
correlation is actually higher for the value-weighted correlation.
We also examine the distribution of the differences between the shares in the two matrices at the
3-digit I-O-commodity/I-O industry-cell level. Figure 1 exhibits the distribution of share
differences for 3-digit cells. Most of the cells have very small differences in the share of the
imported intermediate input (I-O commodity) across industries. This suggests that for most cells,
the share of the commodity imports allocated to a particular industry is fairly close in many cells.
The high correspondence may be due to the large number of cells for which both methods
allocate zero imports.
We also examine the BEA import-value-weighted distribution of share differences to see
whether we match the allocated shares as closely for I-O-commodity/I-O-industry cells with
relatively large import values. Figure 2 shows the BEA-import-value-weighted distribution. The
value-weighted distribution is obviously more dispersed. For I-O-commodity/I-O-industry cells
with relatively high import values, the import matrices derived from the “import comparability
assumption” method and our alternative method are more different. In contrast to the unweighted
198
Characterizing the differences at a detailed level is difficult because of the constraint of confidentiality.
This excludes I-O commodities that the BEA allocates to I-O industries outside of manufacturing. I-O industries
outside of manufacturing can account for a significant share of value of some I-O commodities.
199
367
distribution, significantly less mass is at zero or small share differences. While most valueweighted cells have differences below 50 percentage points, there is significant mass of the
distribution that differs by 10 percentage points or more.
We also thought it would be useful to show the I-O-commodity/I-O-industry cells where the two
methodologies have the largest share differences. We were constrained a bit by the disclosure
prevention protocols but were able to release 15 cells from the 10 largest positive and 10 largest
negative differences. The cells are listed in Table 4.
The results highlight some of the limitations and conceptual differences inherent in our
alternative approach. For example, the I-O-commodity/I-O-industry cell in the first row of the
top panel of Table 4 shows that our methodology allocated a significant share of I-O-commodity
337, Furniture imports, to I-O-industry 337, Furniture. The BEA had a much smaller share of
furniture imports allocated to this I-O industry. The first row of the negative panel shows that the
BEA allocated a large share of furniture imports to I-O-industry 321, Wood products. In fact, the
BEA allocated very little in terms of import value to I-O-industry wood products, but instead
allocated most of the value of furniture imports outside of the manufacturing sector. Our
allocation methodology allocated furniture imports by furniture manufactures to I-O-industry
337, Furniture. In contrast, the BEA allocated the furniture imports to the Construction sector. It
would require additional research, and in the end it might be infeasible, to determine whether the
furniture importers that are furniture manufactures are adding value to the furniture imports or
are merely acting as wholesalers. Yet, to determine which allocation method is more appropriate
would require this type of investigation.
The 3-digit I-O-commodity/I-O-industry cells are a bit unsatisfactory because of the relatively
high level of aggregation (but are necessitated by the disclosure prevention protocols). To
provide some sense of how the matrices compare at a more detailed level, we report descriptive
statistics for weighted and unweighted share differences at the 6-digit I-O-commodity/I-Oindustry level. Table 5 reports the mean share difference (weighted and unweighted) for the I-Ocommodity/I-O-industry cells in the 90–100th percentile and the 0–10th percentile. The
unweighted share differences are relatively small—only about 2 percentage points—at both the
high end and the low end. The BEA-import-value-weighted means tell a somewhat different
story. The weighted-average at the low end is more than 50 percentage points different. There
are some 6-digit cells with relatively large import values where the BEA allocates significantly
more import value to the industry than the alternative measure does. At the other end of the
distribution, the differences are smaller, about 16 percentage points different. While it would be
desirable to provide more information on the share differences at a very detailed level,
fortunately the aggregation to the 3-digit level does not seem to distort the overall story much.
In summary, this alternative methodology seems to offer promise and probably warrants
additional investigation. The comparison of the import matrices derived from the alternative
methodology to BEA’s matrix highlight some of the data limitations confronted by the firm-level
methodology and possibly point out some conceptual differences between the two
methodologies. To resolve which allocation is more appropriate would require additional
information. As such, the exercise points out some potential shortcomings in current data
368
collection systems; of particular interest is additional work to resolve the issue of firms importing
intermediate inputs that are not reported as being used in production.
CONCLUSION
In the first part of this paper, we explored alternatives to the Feenstra-Hanson (1999) measure of
offshoring that still make use of the import comparability assumption. While that measure of
offshoring was intended to reflect imported intermediate inputs, in practice it also included
imported final goods. So in this first step, we recalculated the Feenstra-Hanson (1999) measure
of offshoring while focusing on only imported intermediate inputs as defined by end-use
classifications.
In the second step, we explored a different methodology for allocating imported inputs across
industries using firm-level data on imports and production. We used linked production and
import data to construct firm-level I-O tables and then aggregate these to the industry level to
derive imported input intensity by industry. We compared the results of this alternative allocation
methodology with those obtained by the BEA using the “import comparability” assumption. The
comparison of the import matrices derived from the alternative methodology to the BEA’s
matrix highlight some of the data limitations confronted by the firm-level methodology, and
possibly point out some conceptual differences between the two methodologies.
Table 1
Offshoring Trends with original Feenstra-Hanson calculation (Equation 1)
Year
1980
1990
2000
2006
Narrow
Measure
0.047
0.067
0.103
0.129
Broad Measure
0.071
0.123
0.228
0.282
Broad minus
Narrow Measure
0.024
0.055
0.124
0.152
Table 2
Revised Offshoring Trends (Equation 2)
Year
1980
1990
2000
2006
With Investment Goods Included
Narrow Measure
Broad Measure
0.032
0.066
0.054
0.121
0.091
0.197
0.119
0.270
369
Without Investment Goods
Broad Measure
Narrow Measure
0.032
0.065
0.049
0.122
0.083
0.204
0.105
0.274
Table 3
Correlation between Import Value Share across 3-digit IO-Commodity IO-Industry Cells
Unweighted
Weighted (by BEA import value)
0.6803
0.8717
Table 4
IO Commodity IO Industry Cells with Largest Share Differences
3-digit IO Commodity Group
337 Furniture and Related Products
324 Petroleum and Coal Products
315 Apparel
326 Plastics and Rubber Products
323 Printing and Related Support Activities
316 Leather and Allied Products
325 Chemicals
335 Electrical Equipment and Components
3-digit IO Commodity Group
114 Fishing, Hunting, and Trapping
311 Food
3-digit IO Industry Group
326 Plastics and Rubber Products
334 Computer and Electronic Products
325 Chemicals
335 Electrical Equipment and Components
3-digit IO Industry Group
321 Wood Products
311 Food
312 Beverage and Tobacco Products
325 Chemicals
314 Textile Products
Alt. Share
0.50
0.82
0.46
0.56
0.38
0.61
0.73
0.40
Alt. Share
0.06
0.18
0.12
0.00
0.13
0.00
0.00
BEA Share
0.01
0.34
0.00
0.18
0.01
0.26
0.46
0.20
BEA Share
0.98
1.00
0.73
0.36
0.46
0.22
0.22
Share Difference
0.50
0.48
0.46
0.38
0.37
0.35
0.28
0.20
Share Difference
-0.92
-0.82
-0.62
-0.36
-0.32
-0.22
-0.22
Note: This table lists the 3-digit IO Commodity IO Industry cells with the largest share differences (both positive and negative). The table lists 8 of
the top 10 positive differences and 7 of the top 10 negative differences. The remaining cells were suppressed to prevent disclosure.
Table 5
Mean Differences in Shares for 6-digit level IO-Commodity IO_Industry Cells
0 to 10th Percentile
-0.021
-0.543
Unweighted
Weighted (by BEA import value)
370
90 to 100th Percentile
0.020
0.165
Figure 1
Histogram of Difference in Commodity-Industry Cell Import Share
Figure 2
Histogram of Difference in Commodity-Industry Cell Import Share
(Weighted by BEA Import Value)
371
Appendix A: End-Use Final Goods
Personal Consumption Expenditure:
The following include both final and intermediate goods:
00020
00100
00110
00120
00130
00140
00150
00160
00170
00180
00190
01000
01010
01020
15200
16110
Cane and beet sugar
Meat products & poultry
Dairy products & poultry
Fruits & preparations including juices
Vegetables & preparations
Nuts & preparations
Food oils & oilseeds
Bakery products & confectionery
Tea, spices, & preparations
Agricultural foods, n.e.c.
Wine & related products
Fish and shellfish
Whiskey and other alcoholic beverages
Other nonagricultural foods & food additives
Fabricated metal products
Blank audio and visual tapes and other media
The following are final goods only:
40000
40010
40020
40030
40040
40050
40100
40110
40120
40140
41000
41010
41020
41030
41040
41050
41100
41110
41120
41130
41140
Apparel, & household goods--cotton
Apparel, & household goods--wool
Apparel, & household goods--other textiles
Non-textile apparel & household goods
Footwear of leather, rubber & other materials
Sporting & camping apparel, footwear & gear
Medicinal, dental, & pharmaceutical preparations includ. vitamins
Books, magazines, & other printed matter
Toiletries & cosmetics
Consumer nondurables, n.e.c.
Furniture, household items & baskets
Glassware, porcelain, & chinaware
Cookware, cutlery, house & garden ware & tools
Household and kitchen appliances
Rugs & other textile floor coverings
Other household goods
Motorcycles & parts
Pleasure boats & motors
Toys, shooting & sporting goods, including bicycles
Photographic & optical equipment
Musical instruments & other recreational equipment
373
41200 Television receivers, video receivers, & other video equipment
41210 Radios, phonographs, tape decks, & other stereo equipment & parts
41220 Records, tapes, & disks
413
Coins, gems, jewelry, & collectibles
42000 Unmanufactured goods
421 Unmanufactured diamonds
Investment (final goods):
20000
20005
21000
21100
21110
21120
21130
21140
21150
21160
21170
21180
21190
21200
21400
21500
21600
22000
22010
22020
22100
22200
22300
Generators, transformers, and accessories
Electrical equipment and parts n.e.c.
Oil-drilling, mining, and construction machinery
Industrial engines, pumps, compressors, and generators
Food- and tobacco-processing machinery
Machine tools & metal-working machinery, molding and rolling
Textile, sewing and leather working machinery
Woodworking, glass-working & plastic- and rubber-molding mach.
Pulp & paper machinery, bookbinding, printing & packaging mach.
Measuring, testing, and control instruments
Materials-handling equipment
Other industrial machinery
Photo- & service-industry machinery and trade tools
Agricultural machinery and equipment
Telecommunications equipment
Other business machines
Scientific, hospital, and medical equipment and parts
Civilian aircraft, complete*
Civilian aircraft, parts
Civilian aircraft, engines
Railway & other commercial transportation equipment
Vessels (except military & pleasure craft) & misc. vehicles
Spacecraft, engines & parts, except military
Automotive Vehicles, Parts, and Engines (final and intermediate goods):**
30000 Passenger cars, new and used
30100 Complete and assembled
Raw Materials (not final goods nor intermediate inputs):*
14200
14220
14240
14250
14260
14270
Bauxite and aluminum
Copper
Nickel
Tin
Zinc
Nonmonetary gold
374
14280 Other precious metals
14290 Misc. non-ferrous metals
10
Crude, Fuel oil, Other petroleum products, Coal, Gas, Nuclear fuel, Electric energy
_____________
* These classifications are always excluded from the offshoring calculation.
** This broad category include both final and intermediate goods. Those listed here are final goods and are excluded
from the offshoring calculation.
375
References
Bergstrand, Jeffrey H., and Peter Egger. 2008. “Finding Vertical Multinational Enterprises.”
University of Notre Dame, University of Munich and CESifo, February.
Feenstra, Robert C. 1996. “U.S. Imports, 1972–1994: Data and Concordances.” NBER Working
Paper No. 5515. Cambridge, MA: National Bureau of Economic Research.
Feenstra, Robert C., and Gordon H. Hanson. 1999. “The Impact of Outsourcing and HighTechnology Capital on Wages: Estimates for the U.S., 1979–1990.” Quarterly Journal of
Economics 114(3): 907–940.
Feenstra, Robert C., John Romalis, and Peter Schott. 2002. “U.S. Imports, Exports and Tariff
Data, 1989–2001.” NBER Working Paper No. 9387. Cambridge, MA: National Bureau of
Economic Research.
Houseman, Susan N. 2008. “Outsourcing and Offshoring: Problems for Price and Productivity
Measurement.” W.E. Upjohn Institute for Employment Research, June.
National Research Council. 2006. Analyzing the U.S. Content of Imports and the Foreign
Content of Exports. Washington, DC: National Academies Press.
Sitchinava, Nino. 2007. “Market Structure Index of HTS Imports.” Photocopy, University of
Oregon.
———. 2008. “Trade, Technology, and Wage Inequality: Evidence from U.S. Manufacturing,
1989–2004.” PhD diss., University of Oregon.
Winkler, Deborah, and William Milberg. 2009. “Errors from the ‘Proportionality Assumption’ in
the Measurement of Offshoring: Application to German Labor Demand.” SCEPA
Working Paper 2009-12. New York: Bernard Schwartz Center for Economic Policy
Analysis and Department of Economics, New School for Social Research.
Wright, Greg C. 2009. “Task-Specific Offshoring Costs: Measurement across Countries and
Over Time.” PhD diss., University of California, Davis.
377
Errors from the “Proportionality Assumption” in the
Measurement of Offshoring:
Application to German Labor Demand
Deborah Winkler*
William Milberg**
October 5, 2009
* Research Fellow, Schwartz Center for Economic Policy Analysis, The New School for Social Research, and
Consultant, The World Bank. Email: [email protected].
The author is grateful to the Schwartz Center for Economic Policy Analysis for financial support. The views
expressed in the paper are those of the authors and should not be attributed to the World Bank, its Executive
Directors or the countries they represent.
** Professor, Department of Economics, New School for Social Research and Faculty Research Fellow, Schwartz
Center for Economic Policy Analysis. Email: [email protected].
379
Errors from the “Proportionality Assumption” in the
Measurement of Offshoring: Application to German Labor Demand
Deborah Winkler and William Milberg
ABSTRACT
Offshoring—the importing of intermediate materials and services—has expanded rapidly in most
industrialized countries, and its impact on the labor markets in these countries has been the
source of enormous debate in both scholarly and popular circles. Since data on imported inputs at
the sectoral level are not available for the United States and the United Kingdom, empirical
research has relied entirely on a proxy-based measure offshoring, using what the OECD refers to
as the “proportionality assumption.” That is, every sector is assumed to import inputs of each
material and service in the same proportion as its economy-wide use of that input.
German input-output data differentiate between domestically purchased inputs and imported
inputs, which permits us to calculate a direct measure of sectoral imported input use. In this
paper, we compare this measure to the proxy-based measure based on the standard
proportionality assumption. We find that the direct measure differs significantly from the proxybased measure for both services and materials offshoring. To assess the significance of using
different measures, we substitute them for each other in standard labor demand equations
focusing on German manufacturing between 1995 and 2004. We find that using the direct
measure of offshoring gives very different results for labor demand—sometimes of opposite sign
—compared to estimates using the proxy-based measure.
We perform a simple decomposition of the proxy-based measure and find that it fails to
accurately capture the cross-sectoral variation in offshoring intensity because—as a result of the
proportionality assumption—it is heavily influenced by the cross-sectoral variation in domestic
input demand. The implications of our findings go beyond the case of Germany. They indicate
that researchers must be cautious about drawing policy conclusions from estimates using the
proxy-based measure of offshoring.
JEL No. F1, F2
Key Words: Services Offshoring, Offshoring Intensity, Labor Demand
380
Offshoring—the importing of intermediate materials and services—has expanded rapidly in most
industrialized countries, and its impact on the labor markets in these countries has been the
source of enormous debate in both scholarly and popular circles.200 Since data on imported inputs
at the sectoral level are not available for the United States and the United Kingdom, empirical
research has relied entirely on a proxy-based measure offshoring, using what the OECD refers to
as the “proportionality assumption” (Grossman and Rossi-Hansberg 2006). The U.S. Bureau of
Economic Analysis, for example, collects data on input use, but does not break out imported from
domestically produced inputs. Lacking information on a sector’s imports of each input,
researchers have instead applied the economy-wide import penetration ratio for a material or
service input to approximate the imported input share of that material or service by all sectors.
That is, every sector is assumed to import inputs of each material and service in the same
proportion as its economy-wide use of that input. Without the information on imported input use,
the proportionality assumption has been accepted in most major studies of the level and impact of
offshoring.201
To date, there has been no way to assess the extent of error in measurement introduced by the use
of the proportionality assumption, but recent data for Germany provide a test. German inputoutput data differentiate between domestically purchased inputs and imported inputs, which
permits us to calculate a direct measure of sectoral imported input use. In this paper, we compare
this measure to the proxy-based measure based on the standard proportionality assumption. We
find that the direct measure differs significantly from the proxy-based measure for both services
and materials offshoring. To assess the significance of using different measures, we substitute
them for each other in standard labor demand equations. We find that using the direct measure of
offshoring gives very different results for labor demand—sometimes of opposite sign—
compared to estimates using the proxy-based measure. For example, using the proxy-based
measure, services offshoring is found to have a positive and statistically significant effect on
German employment. Using the direct measure, the estimated employment effect is significantly
negative. This result is robust to a number of specifications and estimation techniques.
We perform a simple decomposition of the proxy-based measure and find that it fails to
accurately capture the cross-sectoral variation in offshoring intensity because—as a result of the
proportionality assumption—it is heavily influenced by the cross-sectoral variation in domestic
input demand. The implications of our findings go beyond the case of Germany. They indicate
that researchers must be cautious about drawing policy conclusions from estimates using the
proxy-based measure of offshoring when we know, at least in the case of Germany for 1995–
2004, that the direct measure gives a very different result.
This paper has five sections. First we discuss the alternative measures of offshoring—the direct
measure and the proxy-based measure—differentiating between services and materials offshoring
intensities, and present our calculations on these two measures for Germany in 1995 and 2004.
We then look at the source of the apparent error in the proxy-based measure, followed by an
200
For a concise survey of the scholarly literature on employment and wage effects of offshoring, see Milberg and
Schöller (2008). For a discussion of the parallels between the scholarly and popular debates, see Milberg (2008).
201
The assumption was first used by Feenstra and Hanson (1996, 1999) and has been adopted in all major studies,
for example, Hummels, Ishi, and Yi (2001), Amiti and Wei (2005, 2006), and Grossman and Rossi-Hansberg
(2006). Further discussion of the proportionality assumption can be found in National Research Council (2006).
381
econometric analysis of offshoring and labor demand that uses the two measures and confirms
the error of the proxy measure. In the final section we conclude with a discussion of the
implications for future research and data collection.
OFFSHORING INTENSITY
In this section, we calculate the direct and proxy measures of services and materials
offshoring intensity for Germany from 1995 to 2004. The analysis uses annual input-output
data from the German Federal Statistical Office (FSO). Input-output tables focus “on the
interrelationships between industries in an economy with respect to the production and uses of
their products and the products imported from abroad. In a table form […] the economy is viewed
with each industry listed across the top as a consuming sector and down the side as a supplying
sector” (United Nations 1999, p. 3). We extract a symmetric 43-sector matrix from the
original input-output tables containing 71 sectors, including all 36 manufacturing sectors in
the original tables and seven of the 27 services sectors. We drop the primary sector (sectors
1–3) and the sectors Mining and Quarrying of the secondary sector (sectors 4–8), as they
generally have little or no offshoring activity. Total nonenergy inputs in Equations. (1) and
(5) contain all 36 material inputs plus the 7 service inputs selected above. For a list of the 43
sectors covered, see Appendix A.
In the following section, we use the term “inputs” when we refer to the supplying sectors. The
selection of the 7 service inputs out of 27 follows the aggregation of Kalmbach et al. (2005)
and includes tradable business activities. Business activities comprise Other business activities
(sector 62), as well as the following 6 sectors: 1) Post and telecommunications; 2) Financial
mediation (except insurance and pension funding); 3) Activities related to financial mediation; 4)
Rental of machinery and equipment; 5) Computer and related activities; and 6) Research and
development (sectors 54, 55, 57, 59–61). We exclude Wholesale, trade, and commission excl.
motor vehicles services from the original definition, since in our view they do not represent
typical offshoring services. Abramovsky, Griffith, and Sako (2004), for instance, classify them as
nonbusiness services. Consumer-related202 and social services203 are also not considered, since
the former in general do not represent typical offshoring services and the latter are not tradable.
Direct and Proxy Offshoring Intensity Measures
In this section we present the two measures of offshoring intensity: a direct offshoring
intensity measure that uses direct information on imported input use and a proxy
offshoring intensity measure that adopts the proportionality assumption that all sectors
import an input at the economy wide rate. We explain the two different concepts by using
the example of services offshoring intensity. These definitions can be applied analogously to
materials offshoring intensity.
The direct services offshoring intensity (DOS) measures the share of imported service inputs s in
total nonenergy inputs used by sector i at time t and is calculated as follows:
202
203
Sectors within the classification of the FSO: 45, 47-53, 56, 58, 69–71.
Sectors within the classification of the FSO: 63–68.
382
(1)
(imported input purchases of service s by sector i)t
(total non - energy inputs used by sector i )t
DOSist =
The direct services offshoring intensity across all service inputs s for sector i at time t is
calculated by taking the sum over all DOSist:
(2)
DOSit = ∑ DOSist
s
The sectoral services offshoring intensity DOSit should not be confused with DOSst, which
represents the average offshoring intensity of a certain service input s across all sectors i. This is
calculated by aggregating the respective DOSist, weighted by total sectoral nonenergy inputs
INP , which is204
(3)
DOS st = ∑ DOSist * ( INPit / INPt )
i
, where INPt = ∑ INPit .
i
Summing DOSst over all service inputs s yields the average services offshoring intensity DOSt
across all sectors i and all inputs s at time t
(4)
DOSt = ∑ DOS st
s
.
The second measure is the proxy services offshoring intensity (POS), which uses a proxy for the
proportion of the imported service input s used in home production, defined as follows (see e.g.,
Feenstra and Hanson [1996]:
⎡ (input purchases of service s by sector i ) ⎤ ⎡
(imports of service s )
⎤
t
t
(5) POSist = ⎢ (total non - energy inputs used by sector i ) ⎥ ⎢ production + imports - exports ⎥
t ⎦⎣
st
st
st ⎦
⎣
The first bracket gives the share of the purchased service input s in total nonenergy inputs for
sector i at time t, which we call the sectoral input share. However, the first ratio does not
distinguish between domestically and foreign purchased inputs. Offshoring focuses solely on
inputs purchased from abroad. Therefore, the second bracket gives an adjustment based on the
share of total imported inputs s (the numerator) in the entire domestic disposability of this input s
(the denominator), where the latter is composed of home production plus imports minus exports
at time t. We call the second bracket of Equation (3) the overall import share.
The proxy services offshoring intensity POSist of service input s in sector i is equal to the product
of the two ratios. The proxy measure is based on the assumption of the same import share of
204
Other authors, e.g., Amiti and Wei (2005, 2006), use sectoral outputs as weights. Using total nonenergy inputs
instead of output results in a more accurate overall offshoring intensity, as it directly refers to the denominator of the
offshoring measure.
383
service input s for all sectors, irrespective of actual sectoral differences. In Germany, for instance,
the overall import share of other business activities was 4.5 percent in 2004. Hence, an import
share of 4.5 percent is assumed for each sector i in the calculation of the sectoral import
intensities for 2004. POSit, POSst and POSt are defined analogously to Equations (2), (3), and (4).
We calculate direct and proxy materials offshoring intensities (DOM and POM) analogously to
the services offshoring intensities.
The definition of offshoring intensity suffers from three related shortcomings; the first two
concern both offshoring intensity measures, whereas the last one only holds for the proxy
offshoring measure. First, the numerator underestimates the actual offshoring values, since
import prices—used here for the calculation of the offshoring measure—are generally lower than
the actual purchase prices of these inputs. Second, total nonenergy inputs only include purchased
inputs, but not self-produced inputs used by sector i, which underestimates the denominator.
Third, the application of the same import share of across all sectors in the proxy offshoring
intensity is not accurate, since not every sector uses imports to the same extent. Thus, the
offshoring intensity cannot be exactly measured (Amiti and Wei 2005).
The first two shortcomings are mutually offsetting and the direct offshoring intensities presents a
good measure for the proportion of imported inputs being used by sector i at time t. However, the
third shortcoming—the proportionality assumption that applies the same import share to all
sectors i in the proxy offshoring intensity measures—could constitute a major problem, since
much of the import-induced cross-sectoral variation gets lost. Because of lack of data on the
direct import of intermediates, the proxy measured is used in all major studies of offshoring.205
German Offshoring Intensity Using Direct and Proxy Measures
Table 1 presents the direct and proxy measures of average services offshoring intensity (weighted
by total nonenergy inputs) for each of the 7 selected service inputs s over all 43 sectors i in 1995
and 2004 as defined in Equation (3). For each service input we also show the (unweighted) mean
and the standard deviation across the 43 sectors. The average services offshoring intensity
measured directly, DOSst, more than doubled from 1.37 percent in 1995 to 2.90 percent in 2004.
At the service level, Computer and related activities grew on average from the third smallest
share of 0.08 percent in 1995 to the fourth largest share of 0.39 percent in 2004. Average
offshoring intensities of Research and development services increased from 0.13 percent in 1995
to 0.35 percent in 2004. Other business activities almost doubled their intensities from 0.53
percent in 1995 to 0.95 percent in 2004. These three service inputs are those that are typically
associated with services offshoring and account for 58 percent of the total DOSt in 2004.
The proxy measures of services offshoring intensity, POSst, are shown at the bottom of Table 1.
They are smaller than the direct measures. Applying the overall import share of a services
category s to all sectors i thus seems to underestimate the real amount of imported service inputs.
Average POSt more than doubled from 0.88 percent in 1995 to 1.80 percent in 2004. Table 1 also
shows that cross-sectoral standard deviations are generally much lower using the proxy measures
compared to the direct measures. The corresponding measures of materials offshoring intensity
205
For example, Feenstra and Hanson (1996, 1999); Amiti and Wei (2005, 2006); Grossman and Rossi-Hansberg
(2006); and OECD (2007).
384
by type of materials input can be found in Appendix B. Note that analogously to subscript s in
Equations (1) to (5), subscript m stands for material inputs. In the case of materials offshoring,
we find the reverse: the proxy measures tend to be higher than the direct measures. Cross-sectoral
standard variations, on the other hand, are higher for the proxy measure than for the direct
measure which we will explain in section.
Table 6: Direct and Proxy Measures of Services Offshoring Intensity by Type of Service Input in Germany,
1995 and 2004
S ervice input s
Rank DOS s 19 9 5 Mean
(weighted
average)
Post and telecommunications
3
0.25%
Financial intermediation
6
0.08%
Activities related to financial intermediation
2
0.31%
Renting of machinery and equipment
7
0.00%
Computer and related activities
5
0.08%
Research and development
4
0.13%
Other business activities
1
0.53%
Total DOS t
1.37%
S ervice input s
Rank POS s 19 9 5
(weighted
average)
3
0.09%
Financial intermediation
5
0.05%
2
0.22%
7
0.00%
6
0.04%
4
0.05%
1
0.42%
Total POSt
0.88%
0.25%
0.06%
0.19%
0.00%
0.13%
0.24%
0.35%
1.23%
Mean
0.10%
0.05%
0.26%
0.00%
0.06%
0.10%
0.45%
1.03%
S td
Dev
Rank DOS s 2 0 0 4
(weighted
average)
Mean
S td
Dev
1.49%
2
0.52%
0.08%
6
0.19%
1.24%
3
0.51%
0.00%
7
0.00%
0.62%
4
0.39%
1.00%
5
0.35%
1.03%
1
0.95%
2.53%
2.90%
S td Rank POS s 2 0 0 4
(weighted
Dev
average)
0.11%
4
0.22%
0.10%
6
0.09%
1.19%
2
0.31%
0.00%
7
0.00%
0.20%
3
0.27%
0.44%
5
0.16%
0.41%
1
0.75%
1.44%
1.80%
0.49%
0.18%
0.80%
0.00%
0.64%
0.64%
0.73%
3.48%
Mean
3.04%
0.10%
4.71%
0.00%
2.07%
2.91%
2.06%
6.98%
S td
Dev
0.21%
0.09%
0.38%
0.00%
0.40%
0.27%
0.74%
2.09%
0.64%
0.16%
1.78%
0.00%
1.42%
1.06%
0.63%
2.78%
Source: Own calculations, Data: FSO, revised input-output tables (1995 and 2004).
Figure 1 presents a plot of the development of the average services and materials offshoring
intensities over all sectors in Germany as defined in Equation (4). The continuous lines represent
the direct measures. Average direct services offshoring intensities DOSt have grown considerably
by on average 8.6 percent per year from 1.4 percent in 1995 to 2.9 percent in 2004, possibly due
to the increased use of ICT. Direct materials offshoring intensities have risen by 8.0 percent per
year from 13.2 percent in 1995 to 23.8 percent in 2004. The relatively strong annual growth rate
of materials offshoring compared to services offshoring is somewhat surprising, as the process of
materials offshoring started earlier and perhaps should have reached its limit. One explanation
would be the collapse of communism in Eastern Europe, which was followed by significant
German foreign direct investment in Central and Eastern Europe, and subsequent wave of reimports back to Germany. Another explanation is the growing reliance on East Asian contract
manufacturers.
385
Figure 5: Offshoring Intensities of Intermediate Inputs in Germany (1995-2004)
3.0%
3.0%
3.0%
2.8%
2.9%
1.7%
1.8%
2003
2004
8.6%
2.5%
2.5%
2.2%
1.9%
OSSt
2.0%
1.5%
1.4%
1.5%
0.9%
2.0%
1.8%
1.0%
0.5%
1.9%
1.0%
1.1%
1.3%
1.9%
8.3%
1.5%
0.0%
1995
1996
1997
1998
CAGR
1999
DOSt
POSt
19.6%
17.6%
20%
OSMt
2002
6.8%
25.5%
24.2% 25.1% 24.8%
25%
10%
2001
28.6%
30%
15%
2000
14.3% 14.6%
13.2%
14.6%
19.6%
20.8% 20.1%
2000
2001
22.0%
23.8%
8.0%
15.5%
14.4% 14.7%
12.2%
5%
0%
1995
CAGR
1996
1997
1998
1999
DOMt
2002
2003
2004
POMt
Source: Own calculations. FSO, revised input-output tables (1995-2004). Weighted average across all sectors i
by total nonenergy inputs at time t.
The dashed lines in Figure 1 represent the average proxy measures POSt and POMt. Average
services offshoring intensities POSt are lower than the corresponding DOSt measures.
Nevertheless, the average annual growth rate is still 8.3 percent over the 1995–2004 period. On
the other hand, the proxy measures of materials offshoring intensity POMt are mostly higher than
the corresponding direct measure DOMt. The POMt variable tracks the constant growth trend of
the DOMt measures with a lower CAGR of 6.8 percent. In sum, there is a clear difference in the
average level and variation between the direct and proxy measures.
386
ERROR IN CAPTURING CROSS-SECTORAL VARIATION USING THE PROXY
MEASURE
Loss of Cross-Sectoral Variation
In this section, we are interested how the proxy measure influences the cross-sectoral variation
of offshoring, i.e., the variation across all sectors considered. In Equation (5), we distinguished
between the “sectoral input share” (first bracket) and the “overall import share” (second bracket).
Accordingly, we can attribute the cross-sectoral variation of the proxy measure in Equation (5) to
the “input-induced variation,” i.e., the variation in the first bracket across all sectors, and the
“import-induced variation,” i.e., the variation in the second bracket across all sectors.
Let us study the import-induced variation of the proxy measure compared to the direct measure
in a first step. Applying the same overall import share for a service input s over all sectors i
constitutes a major loss of cross-sectoral variation, which we will show in the following equation.
Let us assume that a sector i only purchases two service inputs, s1 and s2. Then, the calculation of
POSit for a sector i at time t is given by
⎤
2
⎡ (input purchases of service s1 )it ⎤ ⎡
(imports of service s1 )t
POSit = ∑ POSist = ⎢
⎢
⎥
⎥
s =1
⎣ (total non - energy inputs )it ⎦ ⎢⎣ productions1t + importss1t - exportss1t ⎥⎦
(6)
⎤
⎡ (input purchases of service s2 )it ⎤ ⎡
(imports of service s2 )t
+⎢
⎥
⎥⎢
⎣ (total non - energy inputs )it ⎦ ⎢⎣ productions2t + importss2t - exportss2t ⎥⎦
Now imagine the calculation of POSit for a sector j (with j ≠ i ), which uses the same kinds of
inputs as sector i at time t. We can see from Equation (6) that only the first bracket of each
summand— i.e., the input-induced variation—differs from sector i, while the second bracket of
each summand remains as in sector i— i.e. there is no import-induced variation across sectors. As
a consequence, the cross-sectoral variation in offshoring intensities is solely determined by the
input-induced variation. The application of the proportionality assumption thus lowers the
import-induced cross-sectoral variation.
Influence of Domestically Purchased Inputs
We now analyze the input-induced variation of the proxy measure. As we have shown in the
previous section, the cross-sectoral variation of POSit is only determined by the input-induced
variation because of the proportionality assumption. Note that the term input-induced variation in
opposition to import-induced variation could be misleading, as it includes both the variation of
domestically purchased inputs and the variation of imported inputs of s. In the following section
we show that this can lead to a biased sectoral input share (first bracket in Equation 5), because
the cross-sectoral variation is mainly determined by domestically purchased inputs.
The sectoral services offshoring intensities for the 36 manufacturing sectors using direct and
proxy measures are plotted in Appendix C. One can see that the two measures differ for each
sector. The cross-sectoral standard deviations per year (on bottom) are stronger for the DOSit
387
measures. Consequently, the standard deviation of the DOSst measures is also higher compared to
that for POSst, as already shown in Table 1. The category Other business services, for instance,
shows a standard deviation of 1.03 percent in 1995 using the DOSt measures, while the standard
deviation is only 0.41 percent for the POSt measures.
Similar differences between the direct and the proxy measure can be detected for materials
offshoring intensity. Appendix D shows the sectoral materials offshoring intensities in
manufacturing using both measures. Despite the loss in import-induced variation explained in the
previous section, the cross-sectoral standard deviations of POMit are higher than those for DOMit.
Likewise, the cross-sectoral standard deviations for the POMmt measures are also higher
compared to the DOMmt measures (see Appendix B).
Why are there such differences in the standard deviation across sectors between the direct and
proxy measures? In the following equation, we show the extent to which domestically purchased
inputs as opposed to imported inputs influence the input-induced variation. To do this, we
introduce two domestic outsourcing variables to reflect the amount of home-purchased service
inputs and home-purchased material inputs. The domestic services outsourcing intensity HPS is
calculated as follows:
(7)
HPSist =
(domestic input purchases of service s by sector i )t
(total non - energy inputs used by sector i )t
The domestic services outsourcing intensity HPSit for sector i at time t is calculated by taking the
sum over all HPSist: HPSit = ∑ HPSist . The domestic materials outsourcing intensity HPM is
s
calculated analogously.
According to this definition, summing up Equations (7) and (1) yields the left bracket of Equation
(5), i.e., the sectoral input share. Such domestic outsourcing is fully captured in the proxy
measures, and is plotted in Figure 2 for the period 1995–2004. The average domestic outsourcing
intensities are much higher than the offshoring intensities shown in Figure 2.206 We thus expect
domestic outsourcing to exert a stronger influence on the cross-sectoral input-induced variation.
Between 1995 and 2004, domestic services outsourcing grew at an average rate of 5.3 percent per
annum, (from 16.6 percent to 26.3 percent), while the overall domestic materials outsourcing
intensity grew from 39.7 percent in 1995 to 50.7 percent in 2004, a compound annual growth rate
of 2.8 percent.
The cross-sectoral correlations presented in Appendix E show that the sectoral POSit measures
have a very high correlation with the corresponding domestic services outsourcing intensities
HPSit, which is reflected in an average correlation of 0.9. This means that most of the crosssectoral variation in POSit is in fact determined by the domestic services outsourcing intensity
and not by imported service inputs. Despite the fact that the POMit measure shows an overall
correlation with the domestic materials outsourcing intensity HPMit of almost zero, the sectoral
206
Note that the offshoring and domestic outsourcing measures for materials or services do not sum to 100 percent.
The denominator in both measures is ‘total nonenergy inputs,’ which includes both material and service inputs, while
the numerator includes only services or materials, depending on the measure.
388
data reveal that 22 sectors have a pairwise correlation of more than 50 percent. This is due to the
fact that some sector pairs show positive and others have a negative correlation. We explain these
differences by the fact that the ratio of domestic to imported inputs is much higher in services
than in materials (see Figures 1 and 2). The influence of domestic outsourcing, and thus the error,
would seem to be less severe for materials compared to services.
Figure 6: Domestic Outsourcing Intensity of Intermediate Inputs in Germany
Domestic Outsourcing Intensity
55%
2.8%
50%
50.7%
47.3%
50.6%
45%
40%
35%
46.6%
44.5%
42.6%
47.2%
25.2%
25.9%
43.9%
39.7%
30%
25%
20%
47.2%
16.6%
19.8%
17.7%
17.6%
1996
1997
21.9%
23.3%
24.3%
26.3%
5.3%
15%
1995
CAGR
1998
1999
HPSt
2000
2001
2002
2003
2004
HPMt
Source: Own calculations. Data: FSO, revised input-output tables (1995–2004). Weighted average across all
sectors i by total nonenergy inputs at time t.
We conclude that the input-induced variation is mostly determined by domestically purchased
inputs, which is stronger in the case of service inputs. A high standard deviation of the domestic
outsourcing variables thus influences the variation in the proxy measure. To support our
hypothesis, we present the sectoral domestic outsourcing intensities and their standard deviations
in Appendix E. The standard deviations of HPSit. and HPMit are much higher than the respective
standard deviations of DOSit and DOMit, which implies a strong influence of domestically
purchased inputs on the variation of POSit and POMit. Moreover, the standard deviations of
HPMit (with 36 material inputs) across all sectors are higher than the standard deviations of HPSit
(with 7 service inputs).
To sum up: the use of proxy measures significantly influences the degree of cross-sectoral
variation. First of all, the cross-sectoral variation of the proxy measure is only determined by the
input-induced variation due to the proportionality assumption, since there is by assumption no
cross-sectoral variation in the overall import shares. In general this implies less cross-sectoral
variation in the proxy measure of offshoring. Second, the input-induced variation is to a large
extent determined by domestically purchased inputs, which can have an upward or downward
effect on the cross-sectoral variation, depending on the cross-sectoral variation in domestic input
389
demand compared to that for imported inputs. In our specific case, we detected a lower crosssectoral variation in the proxy measure for services, but a higher cross-sectoral variation for
materials. This indicates a strong upward effect of the input-induced variation for materials on the
cross-sectoral variation.
These two effects lead in general to erroneous measurement, and this error may be particularly
important when the proxy measure is used in cross-sectoral analysis of offshoring. We test this
hypothesis in the next section, where we measure the impact of services offshoring on labor
demand in German manufacturing using both the direct and the proxy offshoring measures.
OFFSHORING AND LABOR DEMAND IN GERMANY
Empirical Model
We use a standard model of labor demand, following the labor demand specification of
Hamermesh (1993). A firm’s linearly homogenous production function F with constant returns to
scale is described as follows:
(8)
Y = F(L, K, S, M, T)
∂F
∂2 F
∂2F
> 0,
< 0,
>0
2
∂x1
∂x1∂x2
∂x1
with x1, x2 = L, K, S, M, T
where labor L, capital K, intermediate services S, intermediate materials M, and technology T are
the input factors. The technology shifter, T=T(OS, OM), is a function of services and materials
offshoring OS and OM.207 T represents a change of the production function due to offshoring.
The corresponding linearly homogeneous cost function, conditional on the level of output Y, is
the following:
(9)
S
M
T
C = C(Y, w, r, p , p , p )
∂C
∂C
> 0,
>0
∂c1
∂c1∂c2
S
M
T
with c1,cx2 = w, r, p , p , p
where w designates wages, r the rental rate on capital, p S , p M , and pT the prices for service,
material, and technology inputs, and Y the constant output.
Using Shephard’s Lemma,208 the conditional labor demand function Ld is derived as follows:
(10)
L* = Ld (Y , w, r , p S , p M , pT )
207
We use OS and OM in the following, when the variables can represent the direct or the proxy measure.
According to Shephard’s Lemma (1953), factor demand is determined by the first partial derivative of the cost
function with respect to the corresponding factor price, regardless of the kind of production function.
208
390
In log-linear form, we can write
S
(11)
M
T
ln Lit = α 0 + ηY ln Yit + η L ln wit + η K ln rit + η S ln p it + ηG ln p it + ηT ln p it
In this form, the equation results in the employment-output elasticity ηY , the price elasticity of
demand for labor ηL , the cross-elasticity of demand for labor due to a change in the rental rate on
capital η K , the cross-elasticities of demand for labor due to a change in input prices for services,
goods, and technology ηS , ηM , and ηT .
Wages are observed directly, but some choices must be made on the specification of the input
prices. The rental rate on capital, r, is assumed to be the same for all companies and to be a
function of time r=f(t). r is not directly included in the estimation model, but will be captured by
adding fixed-year dummies. The input prices for service and material inputs p S and p M can be
subdivided into foreign input prices and domestic input prices (see Winkler [2009]). Following
Amiti and Wei (2005), we use offshoring intensities as an inverse proxy for import prices of
services as well as of materials. The lower the prices of imported services or material inputs, the
higher the offshoring intensities should be. Therefore, we use the offshoring variables as inverse
proxies for imported input prices.
Winkler (2009) uses the previously calculated domestic outsourcing intensities HPS and HPM as
an inverse proxy for the prices of home-purchased service and material inputs. However, these
variables can only be calculated using the domestic input matrices of the input-output tables.
Unlike the offshoring intensity measures, we do not know an alternative proxy measure for
domestic outsourcing intensities. Therefore, we do not include HPS and HPM in the regressions.
This also makes our study more comparable with other studies that do not include domestic
outsourcing intensities (e.g., Amiti and Wei [2005, 2006]). Finally, the input prices pT of the
technology shifter T need to be determined. Since adequate measures for pT are not available
and T=T(OS, OM), we use OS and OM as inverse proxies for the prices of technology pT ,
because falling prices of technology inputs pT are expected to be reflected in higher offshoring
intensities.
Equation (11) thus reduces to
lnLit = α0 + ηY lnYit + ηL lnwit + ηOS lnOSit + ηOM lnOMit + δtDt + εit
Note that OS and OM have two functions in Equation (12). First, they are used as (inverse)
proxies for other foreign input prices, and second, they are used as (inverse) proxies for the
prices of the technology shifter T. Higher output is expected to be associated positively with labor
demand, that is, ηY > 0 . Increasing wages are expected to be associated negatively, that is,
η L < 0 . Concerning OS and OM, their net effects are not unambiguous as noted by Amiti and
Wei (2006). Offshoring can influence employment in at least three ways. First, if input prices p S
and p M fall, i.e., if OS and OM increase, labor is likely to be substituted for imported inputs. We
391
call this the input substitution effect. Analogously, if input prices pT decrease, i.e., if OS and OM
rise, labor is likely to be substituted for technology in what we call the technology substitution
effect. Second, offshoring could augment productivity via T, so that less labor is needed for the
same amount of output (productivity effect). The substitution effect influences labor demand in a
direct manner, whereas the productivity effect is indirect.
Besides these two negative effects, scale effects could influence labor demand positively. If
productivity effects lead to lower prices, this would be expected to be associated with a greater
quantity demanded, in turn increasing the demand for labor. Thus, the net effect of offshoring is
not clear. If the negative substitution or productivity effects are larger than the positive scale
effects, then ηOS < 0 and ηOM < 0. If the scale effects dominate the other effects for all variables,
we would expect ηOS > 0 and ηOM > 0
Estimation Results
We estimate the effect of offshoring on labor demand using the consistent fixed effects estimator,
which allows unobserved time-constant sector-specific effects ci to be correlated with some
explanatory variables xit . All estimations produce standard errors robust to both
heteroscedasticity (Huber-White sandwich estimators) and any form of intracluster correlation.
Table 2 shows the results using the fixed effects estimator including all sectors, Table 3 shows
the results excluding outliers ‘pharmaceuticals’ and ‘recycling,’ and Table 4 applies the
instrumental variables two-stage least squares (IV 2SLS) estimator to control for potential
endogeneity of the offshoring variables. The correlation matrix, summary statistics and data
description can be found in Appendices G–I..
392
Table 7: Fixed Effects Estimations (1995–2004)
Dependent variable: lnLt
Fixed effects using DOS and DOM measures
(1)
lnYt
(6)
(7)
(8)
0.2271**
0.1401**
0.3398***
0.1291**
0.1771**
0.0973
(0.020)
(0.050)
(0.032)
(0.005)
(0.028)
(0.036)
(0.203)
0.2484***
0.0613
0.2277**
0.0503
(0.010)
(0.470)
(0.014)
(0.402)
-0.5369***
-0.4414***
-0.4701***
-0.3499***
-0.5357***
-0.4395***
-0.4623***
-0.3621***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.1250**
-0.1560***
-0.1300*
-0.1490***
(0.037)
(0.005)
(0.053)
(0.008)
-0.0403*
-0.0092
-0.0133
0.0121
0.0910*
0.0569
0.0860**
0.0577
(0.059)
(0.653)
(0.313)
(0.549)
(0.064)
(0.133)
(0.030)
(0.132)
-0.0406**
-0.0243**
(0.013)
-0.0068
(0.020)
0.0075
(0.825)
(0.770)
0.0095
-0.0051
0.0258
0.0116
-0.0128
-0.0026
0.0524
0.0444
(0.672)
(0.769)
(0.367)
(0.596)
(0.844)
(0.955)
(0.253)
(0.327)
lnOMt-1
0.0109
(0.665)
ln(IM/Y)t
Yes
Yes
0.0376
0.0160
(0.137)
(0.737)
0.0616
(0.108)
0.0209
0.0429
0.0195
0.0357
(0.602)
(0.127)
(0.449)
(0.194)
ln(IM/Y)t-1
Year fixed effects
Fixed effects using POS and POM measures
(5)
0.1216**
lnOSt-1
lnOMt
(4)
(0.002)
lnwt-1
lnOSt
(3)
0.3418***
lnYt-1
lnwt
(2)
Yes
-0.0396
-0.0368
(0.295)
(0.235)
Yes
Yes
Yes
Yes
Yes
Joint significance tests:
lnYt + lnYt-1 = 0
p>F=0.0062
p>F=0.0970
p>F=0.0181
p>F=0.1650
lnwt + lnwt-1 = 0
p>F=0.0000
p>F=0.0000
p>F=0.0000
p>F=0.0000
lnOSt + lnOSt-1 = 0
p>F=0.0415
p>F=0.0641
p>F=0.2827
p>F=0.2751
lnOGt + lnOGt-1 = 0
p>F=0.8727
p>F=0.3239
p>F=0.9273
p>F=0.1763
ln(IM/Y)t+ln(IM/Y)t-1 = 0
p>F=0.1973
p>F=0.3325
AIC
-855.8
-821.6
-849.8
-809.7
-834.9
-813.5
-879.3
-836.6
Observations
347
312
319
287
360
324
330
297
R-squared
0.64
0.65
0.66
0.66
0.62
0.63
0.68
0.68
Source: Own calculations. p*<0.1, p**<0.05, p***<0.001 (p-values in parentheses).
In each case we consider instantaneous effects and additional one-period lags of the independent
variables. Employment is associated with income and wages in the predicted fashion under all
estimation techniques, whether the estimation includes proxy or direct measures of offshoring. In
all cases the income variable (contemporaneous and one-year lag) is positive and in most cases
statistically significant. Similarly, the wage variable (contemporaneous and one-year lag) is
always negative and in most cases significant. When lagged values of these variables were
included, they were in all cases jointly significant with the contemporaneous value (see joint
significance tests at the bottom of each table).
393
There are different results, however, for the offshoring variables depending on if they are
measured in a direct or proxy fashion. In the fixed effects models (both with and without
outliers), materials offshoring varies from positive to negative but is statistically insignificant in
all models. In the IV 2SLS estimated with year fixed effects without outliers (Table 4), the direct
measure of materials offshoring is negative and significant in columns (1) and (3) and
insignificant in (2) and (4). The proxy measure of offshoring has a negative sign in all cases, and
the effect is larger and statistically significant at a higher level in columns (5) and (7).
Table 8: Fixed Effects Estimations without Outliers (1995–2004)
Fixed effects w/o outliers1) using DOS and DOM
Fixed effects w/o outliers2) using POS and POM
measures
lnYt
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.3547***
0.1297**
0.2578**
0.1508**
0.2717***
0.1210**
0.1716**
0.0983
(0.001)
(0.018)
(0.037)
(0.039)
(0.008)
(0.046)
(0.041)
(0.213)
lnYt-1
lnwt
0.0873
0.1628**
0.0406
(0.009)
(0.334)
(0.029)
(0.503)
-0.4770***
-0.5007***
-0.3831***
-0.4770*** -0.4238***
-0.4959***
-0.3992***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.1090*
-0.1477***
-0.0704
-0.1399**
(0.067)
(0.008)
(0.266)
(0.017)
-0.0397*
-0.0097
-0.0139
0.0112
0.0591
0.0381
0.1102**
0.0744*
(0.056)
(0.641)
(0.292)
(0.589)
(0.114)
(0.206)
(0.020)
(0.056)
lnOSt-1
lnOMt
0.2530***
-0.5627***
lnwt-1
lnOSt
measures
(1)
-0.0398**
-0.0245**
-0.0106
0.0247
(0.013)
(0.023)
(0.728)
(0.297)
0.0065
-0.0040
0.0160
0.0071
-0.0036
0.0060
0.0726
0.0708
(0.788)
(0.825)
(0.629)
(0.780)
(0.957)
(0.902)
(0.109)
(0.112)
lnOMt-1
0.0039
(0.882)
ln(IM/Y)t
0.0263
0.0133
(0.343)
(0.775)
0.0558
(0.128)
0.0388
0.0488
0.0368
0.0470*
(0.398)
(0.136)
(0.112)
(0.054)
ln(IM/Y)t-1
-0.0231
-0.0281
(0.560)
Year fixed effects
Yes
Yes
Yes
Yes
(0.385)
Yes
Yes
Yes
Yes
lnYt + lnYt-1 = 0
p>F=0.0047
p>F=0.1121
p>F=0.0316
p>F=0.1960
lnwt + lnwt-1 = 0
p>F=0.0000
p>F=0.0000
p>F=0.0000
p>F=0.0000
lnOSt + lnOSt-1 = 0
p>F=0.0398
p>F=0.0708
p>F=0.3666
p>F=0.1098
lnOMt + lnOMt-1 = 0
p>F=0.9643
p>F=0.6074
p>F=0.9588
p>F=0.0902
ln(IM/Y)t+ln(IM/Y)t-1 = 0
p>F=0.2608
p>F=0.1069
AIC
-836.8
-803.3
-828.5
-786.3
-897.0
-845.5
-868.7
-828.5
Observations
337
303
309
278
340
306
320
288
R-squared
0.65
0.66
0.67
0.66
0.68
0.67
0.70
0.70
394
1) Columns 1–4 exclude the outlier ‘pharmaceuticals’.
2) Columns 5–8 exclude the outliers ‘pharmaceuticals’ and ‘recycling.’
Table 4: IV 2SLS Fixed Effects Estimations without Outliers (1995-2004)
Instrumental Variables 2SLS: Fixed effects
Instrumental Variables 2SLS: Fixed effects
w/o outlier1) using DOS and DOM measures
w/o outliers2) using POS and POM measures
(1)
lnYt
(2)
lnOMt
Year fixed effects
(5)
(6)
(7)
(8)
0.2368**
0.2549***
0.1439*
0.2825***
0.1893*
0.2545***
0.1083
(0.001)
(0.048)
(0.002)
(0.059)
(0.003)
(0.082)
(0.001)
(0.161)
0.2424***
0.2361***
0.0676
0.1395**
(0.001)
(0.006)
(0.276)
(0.023)
-0.5509***
-0.5124*** -0.4750***
-0.4573*** -0.5419***
-0.4597***
-0.4760***
-0.4261***
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.1405**
-0.1078*
-0.1336*
-0.0722
(0.029)
(0.073)
(0.070)
(0.267)
lnwt-1
lnOSt
(4)
0.3537***
lnYt-1
lnwt
(3)
(0.000)
-0.0511**
-0.0298
-0.0625***
-0.0402
0.0911
0.1872
0.0439
0.1176
(0.021)
(0.365)
(0.008)
(0.161)
(0.106)
(0.186)
(0.494)
(0.468)
-0.0815*
0.0376
-0.1181**
-0.0238
-0.2426***
-0.0216
-0.2201***
-0.0185
(0.063)
(0.430)
(0.019)
(0.627)
(0.001)
(0.884)
(0.005)
(0.901)
No
Yes
No
Yes
No
Yes
No
Yes
lnYt + lnYt-1 = 0
p>F=0.0002 p>F=0.0128
p>F=0.0040 p>F=0.0483
lnwt + lnwt-1 = 0
p>F=0.0000 p>F=0.0000
p>F=0.0000 p>F=0.0000
First stage results:
Shea Partial R-squared:
lnOSt
lnOMt
3)
0.5297
0.5092
0.5274
0.5066
0.3297
0.1923
0.3111
0.1802
0.4320
0.3142
0.4144
0.3173
0.4501
0.3247
0.4405
0.3274
2
2
2
2
2
2
Hanson J statistic P-val. Χ (4)=0.05
Χ (4)=0.49
Χ (4)=0.05
Χ (4)=0.42
Χ (4)=0.04
Χ (4)=0.09
Χ (4)=0.03
Χ2(4)=0.13
AIC
-603.5
-649.6
-610.7
-666.8
-648.1
-692.8
-653.2
-711.2
Observations
235
235
235
235
238
238
238
238
R-squared
0.48
0.59
0.50
0.63
0.51
0.61
0.53
0.65
1) Columns 1–4 exclude the outlier ‘pharmaceuticals.’
2) Columns 5–8 exclude the outliers ‘pharmaceuticals’ and ‘recycling.’
3) Overidentification test of all instruments.
395
2
The dramatic differences in the results between proxy and direct measures occur with services
offshoring. In the fixed effect estimations (Table 2), the proxy variable has a positive and
statistically significant coefficient in columns (5) and (7). The direct services offshoring measure
has a negative sign in all cases and is significant in contemporaneous form in model (1) and in
lagged form in models (2) and (4). A very similar result occurs when outliers are removed (Table
3). In the IV 2SLS fixed effects estimations (Table 4), the direct services offshoring variable is
negative and significant when year fixed effects are not included (columns 1 and 3). By
comparison, the coefficient on the proxy measure of services offshoring (models 5-8) is always
positive and statistically insignificant.
CONCLUDING REMARKS
The proportionality assumption in the measurement of offshoring has been adopted in all the
major empirical, input-output-based studies of offshoring. In this paper we provide a first
assessment of the merits of that assumption. Since Germany collects imported inputs directly, we
were able to construct a direct measure of offshoring to compare to the proxy measure, where the
proxy measure was constructed with the proportionality assumption. We estimated the effect of
offshoring on German labor demand for a sample of 36 sectors over the period 1995–2004 and
found that the direct and proxy measures of offshoring give very different results, especially in
the case of services offshoring. In many cases where the proxy measure gives a positive and
insignificant coefficient, the direct measure has a negative and significant coefficient. This
finding is robust to different estimation techniques.
We also performed a simple decomposition of the proxy-based measure. We find that the proxy
measure fails to accurately capture the cross-sectoral variation in offshoring intensity because—
as a result of the proportionality assumption—it is heavily influenced by the cross-sectoral
variation in domestic input demand. More precisely, the cross-sectoral variation of the proxy
measure is only determined by the input-induced variation, and the input-induced variation is to a
large extent determined by domestically purchased inputs.
The implications of our findings go beyond the case of Germany. Researchers must be cautious
about drawing policy conclusions from estimates using the proxy-based measure of offshoring
when we know, at least in the case of Germany for 1995–2004, that using a direct measure
sometimes gives the opposite result. Whereas the proxy measure would support the view that
workers should have no “fear of offshoring,” the direct measure indicates that offshoring has a
negative effect on labor demand. The two results would support very different policy
prescriptions. Researchers relying (because of a lack of data) on the proxy measure should be
very cautious when interpreting the results of their analysis.
Given our results, we would urge that industrialized countries seek to improve data on imported
intermediates along the lines suggested by Sturgeon (2006). This would be especially important
for the United States, the United Kingdom, and Japan, where offshoring levels and their labor
market effects are known to be significant and the subject of considerable policy debate.
396
Appendices
Appendix A: Sectoral Classification
Manufacturing Sectors (36 Sectors)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
Food products
Beverages
Tobacco products
Textiles
Wearing apparel, dressing and dying of fur
Leather, leather products and footwear
Wood and products of wood and cork
Pulp and paper
Paper products
Publishing
Printing
Coke, refined petroleum products and nuclear fuel
Pharmaceuticals
Chemicals exluding pharmaceuticals
Rubber products
Plastic products
Glass and glass products
Ceramic goods and other non-metallic mineral products
Iron and steel
Non-ferrous metals
Metal castings
Fabricated metal products, except machinery and equipment
Machinery and equipment, n.e.c.
Office, accounting and computing machinery
Electrical machinery and apparaturs, n.e.c.
Radio, television and communication equipment
Medical, precision and optical instruments
Motor vehicles, trailers and semi-trailers
Other transport equipment
Manufacturing n.e.c.
Recycling
Electricity, steam and hot water supply
Gas and gas supply
Collection, purification and distribution of water
Construction site and civil engineering
Construction installation and other construction
Service Sectors (7 Sectors)
37
38
39
40
41
42
43
Financial intermediation except insurance and pension funding
397
Appendix B: Materials Offshoring Intensities per Input Category in Germany
Mate rial input m
DOMm 1995
(weighted average)
Me an
Std De v
DOMm 2004
(weighted average)
Me an
Std De v
POMm 1995
(weighted average)
Me an
Std De v
POMm 2004
(weighted average)
Me an
Std De v
Food products
0.51%
0.35%
1.75%
0.68%
0.60%
2.73%
0.42%
0.30%
1.52%
0.63%
0.43%
2.17%
Beverages
0.02%
0.04%
0.24%
0.03%
0.04%
0.22%
0.02%
0.06%
0.36%
0.04%
0.09%
0.57%
T obacco products
0.01%
0.14%
0.93%
0.00%
0.05%
0.32%
0.00%
0.06%
0.36%
0.00%
0.05%
0.34%
T extiles
0.46%
1.14%
4.18%
0.44%
1.11%
3.69%
0.68%
1.92%
7.36%
0.70%
2.05%
7.44%
Wearing apparel, dressing, and dying of fur
0.17%
0.58%
3.78%
0.17%
0.58%
3.76%
0.14%
0.54%
3.53%
0.16%
0.68%
4.43%
Leather, leather products, and footwear
0.10%
0.86%
5.46%
0.10%
0.73%
4.52%
0.12%
0.91%
5.61%
0.15%
1.12%
6.80%
0.37%
0.46%
1.60%
0.38%
0.48%
1.55%
0.38%
0.52%
2.03%
0.37%
0.53%
1.96%
Pulp and paper
0.67%
1.41%
4.94%
0.83%
1.90%
6.83%
0.76%
1.67%
5.72%
0.91%
1.99%
6.72%
Paper products
0.08%
0.14%
0.29%
0.16%
0.26%
0.43%
0.12%
0.25%
0.66%
0.30%
0.54%
1.35%
Publishing
0.03%
0.06%
0.28%
0.11%
0.16%
0.75%
0.02%
0.02%
0.05%
0.11%
0.11%
0.34%
Printing
0.04%
0.04%
0.10%
0.06%
0.06%
0.12%
0.08%
0.13%
0.38%
0.07%
0.12%
0.34%
Coke, refined petroleum products, and nuclear fuel
0.38%
0.60%
2.19%
0.78%
1.17%
3.94%
0.31%
0.60%
2.47%
0.75%
1.41%
6.26%
Pharmaceuticals
0.11%
0.21%
1.29%
0.18%
0.35%
2.25%
0.11%
0.13%
0.77%
0.52%
0.88%
5.75%
1.93%
1.76%
3.60%
3.39%
3.58%
6.63%
2.22%
2.03%
3.18%
3.88%
3.27%
5.46%
Rubber products
0.21%
0.15%
0.48%
0.40%
0.32%
1.16%
0.21%
0.17%
0.49%
0.50%
0.39%
1.31%
Plastic products
0.44%
0.36%
0.46%
0.77%
0.67%
0.91%
0.56%
0.45%
0.80%
0.94%
0.74%
1.31%
Ceramic goods & other non-metallic mineral
products
Iron and steel
0.15%
0.36%
1.75%
0.23%
0.50%
2.55%
0.16%
0.33%
1.31%
0.25%
0.50%
2.04%
0.35%
0.27%
0.75%
0.34%
0.31%
0.73%
0.47%
0.41%
1.52%
0.38%
0.36%
1.27%
1.02%
0.57%
1.48%
1.68%
1.27%
4.32%
1.17%
0.73%
3.03%
1.83%
1.12%
4.52%
Non-ferrous metals
1.10%
1.19%
4.98%
1.69%
2.63%
9.80%
1.11%
1.37%
5.02%
1.78%
2.64%
9.88%
Metal castings
0.07%
0.03%
0.12%
0.26%
0.23%
1.25%
0.10%
0.05%
0.11%
0.29%
0.13%
0.34%
Fabricated metal products, excl. machinery & equip.
0.52%
0.38%
0.51%
1.00%
0.71%
0.99%
0.67%
0.44%
0.76%
1.24%
0.78%
1.41%
0.78%
0.46%
1.08%
1.84%
1.05%
2.47%
0.96%
0.62%
1.15%
1.93%
1.20%
2.39%
Office, accounting, and computing machinery
0.30%
0.52%
2.56%
0.59%
1.05%
4.16%
0.39%
0.71%
2.85%
1.06%
1.94%
7.60%
0.70%
0.37%
1.04%
1.52%
0.80%
2.48%
1.02%
0.57%
1.63%
2.14%
1.20%
3.65%
Radio, television, and communication equipment
0.61%
0.59%
2.46%
1.67%
1.53%
5.98%
0.60%
0.61%
2.14%
2.20%
2.36%
8.93%
Medical, precision, and optical instruments
0.14%
0.20%
0.65%
0.33%
0.48%
1.59%
0.16%
0.27%
0.75%
0.37%
0.64%
2.12%
Motor vehicles, trailers, and semi-trailers
1.52%
0.26%
1.57%
2.83%
0.56%
2.98%
1.12%
0.19%
1.08%
3.86%
0.64%
3.45%
0.11%
0.15%
1.01%
0.41%
0.55%
3.60%
0.11%
0.14%
0.90%
0.70%
0.95%
6.21%
0.09%
0.12%
0.63%
0.30%
0.40%
2.52%
0.07%
0.09%
0.49%
0.24%
0.26%
1.49%
Recycling
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
Electricity, steam, and hot water supply
0.03%
0.06%
0.15%
0.48%
0.91%
5.31%
0.03%
0.06%
0.08%
0.24%
0.42%
0.83%
Gas and gas supply
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
Collection, purification, and distribution of water
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.01%
0.01%
0.01%
0.00%
0.01%
0.02%
0.01%
0.01%
0.03%
0.00%
0.00%
0.01%
Construction installation & other construction
0.14%
0.08%
0.32%
0.21%
0.12%
0.55%
0.01%
0.01%
0.02%
0.02%
0.02%
0.04%
Total Mate rials O ffshoring Inte nsity in t
13.17%
13.91% 10.86%
23.84%
25.19% 16.90%
Source: Source: Own illustration. Data: FSO, revised input-output tables (1995 and 2004).
398
14.31%
16.36% 12.69%
28.57%
29.57% 19.68%
Appendix C: Sectoral DOS vs. POS Measures
DOSit
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t 1995
POSit
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t
Food products
0.0011 0.0008 0.0009 0.0009 0.0010 0.0020 0.0025 0.0029 0.0031 0.0032
0.0010
0.0059 0.0062 0.0070 0.0077 0.0085 0.0100 0.0118 0.0113 0.0101 0.0103
0.0021
Beverages
0.0023 0.0038 0.0041 0.0045 0.0042 0.0051 0.0070 0.0070 0.0052 0.0050
0.0014
0.0134 0.0146 0.0154 0.0164 0.0180 0.0206 0.0216 0.0202 0.0172 0.0165
0.0027
Tobacco products
0.0081 0.0134 0.0150 0.0172 0.0193 0.0226 0.0331 0.0351 0.0376 0.0391
0.0112
0.0170 0.0072 0.0197 0.0220 0.0126 0.0303 0.0323 0.0293 0.0246 0.0269
0.0081
Textiles
0.0002 0.0005 0.0003 0.0002 0.0002 0.0013 0.0016 0.0024 0.0014 0.0015
0.0008
0.0030 0.0030 0.0033 0.0037 0.0038 0.0043 0.0047 0.0050 0.0039 0.0041
0.0007
0.0000 0.0003 0.0003 0.0008 0.0008 0.0011 0.0014 0.0015 0.0009 0.0009
0.0005
0.0027 0.0028 0.0031 0.0031 0.0033 0.0034 0.0039 0.0040 0.0029 0.0032
0.0004
0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.0000
0.0017 0.0017 0.0021 0.0020 0.0020 0.0023 0.0029 0.0027 0.0020 0.0021
0.0004
0.0006 0.0051 0.0051 0.0049 0.0054 0.0015 0.0020 0.0019 0.0013 0.0013
0.0019
0.0034 0.0034 0.0036 0.0040 0.0049 0.0053 0.0052 0.0047 0.0037 0.0039
0.0008
Pulp and paper
0.0020 0.0014 0.0011 0.0011 0.0011 0.0039 0.0045 0.0047 0.0043 0.0043
0.0016
0.0024 0.0023 0.0027 0.0031 0.0036 0.0048 0.0047 0.0047 0.0040 0.0042
0.0010
Paper products
0.0040 0.0013 0.0012 0.0012 0.0012 0.0066 0.0060 0.0061 0.0078 0.0077
0.0029
0.0025 0.0024 0.0028 0.0031 0.0034 0.0044 0.0046 0.0046 0.0040 0.0044
0.0009
Publishing
0.0064 0.0105 0.0124 0.0119 0.0118 0.0229 0.0283 0.0271 0.0287 0.0302
0.0092
0.0165 0.0187 0.0211 0.0250 0.0277 0.0302 0.0322 0.0312 0.0278 0.0294
0.0055
Printing
0.0025 0.0026 0.0027 0.0032 0.0033 0.0050 0.0058 0.0066 0.0053 0.0055
0.0015
0.0029 0.0031 0.0034 0.0042 0.0048 0.0053 0.0059 0.0062 0.0050 0.0054
0.0012
0.0035 0.0024 0.0024 0.0020 0.0030 0.0037 0.0050 0.0062 0.0068 0.0068
0.0019
0.0147 0.0151 0.0159 0.0126 0.0148 0.0100 0.0097 0.0118 0.0136 0.0140
0.0021
Pharmaceuticals
0.0691 0.0714 0.0916 0.0704 0.0923 0.1477 0.1499 0.1666 0.2091 0.1965
0.0540
0.0146 0.0182 0.0232 0.0318 0.0357 0.0450 0.0526 0.0576 0.0613 0.0577
0.0175
0.0037 0.0030 0.0038 0.0041 0.0046 0.0093 0.0120 0.0126 0.0158 0.0164
0.0053
0.0047 0.0051 0.0058 0.0065 0.0074 0.0093 0.0095 0.0095 0.0081 0.0088
0.0018
Rubber products
0.0027 0.0041 0.0043 0.0063 0.0065 0.0078 0.0089 0.0091 0.0068 0.0070
0.0021
0.0046 0.0049 0.0058 0.0072 0.0077 0.0097 0.0097 0.0085 0.0075 0.0080
0.0018
Plastic products
0.0095 0.0054 0.0060 0.0062 0.0066 0.0153 0.0167 0.0183 0.0122 0.0129
0.0048
0.0053 0.0059 0.0070 0.0079 0.0095 0.0116 0.0126 0.0129 0.0096 0.0105
0.0027
0.0016 0.0040 0.0049 0.0052 0.0059 0.0052 0.0056 0.0068 0.0053 0.0060
0.0014
0.0088 0.0088 0.0101 0.0108 0.0120 0.0143 0.0149 0.0142 0.0125 0.0131
0.0023
Ceramic goods & other non-metallic mineral products
0.0087 0.0063 0.0062 0.0062 0.0063 0.0102 0.0126 0.0128 0.0099 0.0096
0.0026
0.0102 0.0103 0.0111 0.0119 0.0138 0.0150 0.0148 0.0141 0.0119 0.0122
0.0018
Iron and steel
0.0012 0.0007 0.0007 0.0008 0.0007 0.0027 0.0032 0.0028 0.0063 0.0068
0.0023
0.0015 0.0016 0.0019 0.0022 0.0022 0.0028 0.0032 0.0028 0.0027 0.0031
0.0006
Non-ferrous metals
0.0014 0.0004 0.0007 0.0008 0.0008 0.0068 0.0080 0.0082 0.0070 0.0070
0.0035
0.0014 0.0015 0.0018 0.0022 0.0025 0.0036 0.0038 0.0036 0.0030 0.0033
0.0009
Metal castings
0.0002 0.0016 0.0015 0.0030 0.0028 0.0026 0.0034 0.0039 0.0057 0.0063
0.0019
0.0026 0.0026 0.0033 0.0042 0.0046 0.0055 0.0056 0.0060 0.0060 0.0070
0.0015
0.0014 0.0024 0.0027 0.0026 0.0028 0.0036 0.0048 0.0052 0.0043 0.0044
0.0012
0.0037 0.0041 0.0045 0.0054 0.0061 0.0069 0.0076 0.0073 0.0068 0.0071
0.0014
0.0032 0.0027 0.0031 0.0030 0.0031 0.0054 0.0062 0.0066 0.0057 0.0059
0.0016
0.0044 0.0051 0.0056 0.0066 0.0073 0.0087 0.0101 0.0099 0.0094 0.0098
0.0022
0.0044 0.0196 0.0289 0.0363 0.0303 0.0270 0.0367 0.0362 0.0263 0.0224
0.0098
0.0064 0.0072 0.0096 0.0117 0.0112 0.0174 0.0202 0.0183 0.0155 0.0136
0.0047
0.0052 0.0049 0.0054 0.0059 0.0053 0.0085 0.0091 0.0112 0.0091 0.0098
0.0023
0.0050 0.0057 0.0064 0.0078 0.0084 0.0085 0.0116 0.0123 0.0110 0.0119
0.0027
0.0060 0.0025 0.0027 0.0025 0.0037 0.0128 0.0162 0.0174 0.0120 0.0112
0.0059
0.0056 0.0063 0.0071 0.0072 0.0097 0.0136 0.0161 0.0159 0.0139 0.0133
0.0041
0.0043 0.0044 0.0050 0.0052 0.0050 0.0079 0.0090 0.0094 0.0079 0.0076
0.0020
0.0055 0.0061 0.0070 0.0087 0.0097 0.0121 0.0133 0.0133 0.0121 0.0117
0.0030
0.0027 0.0008 0.0009 0.0011 0.0013 0.0061 0.0067 0.0065 0.0061 0.0065
0.0027
0.0019 0.0025 0.0027 0.0037 0.0047 0.0062 0.0066 0.0061 0.0045 0.0049
0.0017
0.0013 0.0026 0.0032 0.0038 0.0036 0.0036 0.0040 0.0043 0.0039 0.0038
0.0009
0.0038 0.0055 0.0059 0.0069 0.0075 0.0098 0.0124 0.0123 0.0101 0.0098
0.0029
0.0001 0.0003 0.0003 0.0003 0.0003 0.0017 0.0022 0.0025 0.0018 0.0018
0.0009
0.0047 0.0049 0.0055 0.0063 0.0069 0.0082 0.0087 0.0080 0.0068 0.0072
0.0014
Recycling
0.0000 0.0049 0.0061 0.0067 0.0049 0.0012 0.0014 0.0011 0.0013 0.0019
0.0024
0.0029 0.0033 0.0049 0.0054 0.0050 0.0092 0.0107 0.0098 0.0088 0.0110
0.0031
0.0037 0.0059 0.0068 0.0075 0.0068 0.0069 0.0076 0.0064 0.0092 0.0100
0.0017
0.0080 0.0090 0.0101 0.0122 0.0134 0.0133 0.0154 0.0121 0.0099 0.0105
0.0023
Gas and gas supply
0.0015 0.0077 0.0092 0.0098 0.0127 0.0071 0.0090 0.0098 0.0093 0.0105
0.0029
0.0073 0.0071 0.0093 0.0110 0.0166 0.0222 0.0240 0.0221 0.0200 0.0224
0.0068
0.0000 0.0070 0.0081 0.0086 0.0092 0.0054 0.0061 0.0059 0.0043 0.0049
0.0026
0.0066 0.0074 0.0080 0.0102 0.0118 0.0116 0.0115 0.0111 0.0091 0.0104
0.0019
0.0015 0.0031 0.0032 0.0027 0.0026 0.0017 0.0021 0.0024 0.0018 0.0018
0.0006
0.0052 0.0059 0.0065 0.0067 0.0075 0.0082 0.0077 0.0070 0.0064 0.0063
0.0009
0.0009 0.0011 0.0011 0.0010 0.0010 0.0014 0.0018 0.0022 0.0017 0.0017
0.0004
0.0033 0.0038 0.0039 0.0046 0.0047 0.0052 0.0055 0.0058 0.0046 0.0049
0.0008
Cross-sectoral Std. Dev.
0.0113 0.0119 0.0155 0.0127 0.0157 0.0243 0.0251 0.0277 0.0345 0.0324
Source: Own illustration. Data: FSO, revised input-output tables (1995–2004).
399
0.0043 0.0044 0.0055 0.0065 0.0070 0.0089 0.0100 0.0103 0.0105 0.0101
Appendix D: Sectoral DOM vs. POM Measures
DOMit
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t 1995
POMit
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t
Food products
0.1416 0.1221 0.1442 0.1181 0.1169 0.1479 0.1711 0.1660 0.2191 0.2392
0.0418
0.1317 0.1425 0.1771 0.1428 0.1383 0.1615 0.1725 0.1728 0.1912 0.2016
0.0238
Beverages
0.0931 0.1165 0.1184 0.1329 0.1308 0.1058 0.1114 0.1101 0.1115 0.1058
0.0118
0.1103 0.1087 0.1187 0.1099 0.1114 0.1334 0.1340 0.1326 0.1486 0.1461
0.0154
Tobacco products
0.1208 0.0873 0.0843 0.0720 0.1112 0.0838 0.1087 0.1058 0.0807 0.0872
0.0161
0.0945 0.0755 0.0738 0.0686 0.1343 0.0903 0.1029 0.1042 0.0950 0.1085
0.0195
Textiles
0.3591 0.2665 0.2810 0.3012 0.2924 0.3860 0.4141 0.3915 0.3689 0.3600
0.0521
0.4638 0.4446 0.4863 0.4673 0.4657 0.4923 0.4863 0.4859 0.4875 0.4798
0.0150
0.4549 0.4785 0.5779 0.6177 0.4514 0.4739 0.5755 0.5617 0.4205 0.4599
0.0686
0.6082 0.5570 0.6823 0.7069 0.5990 0.6789 0.6677 0.6891 0.6452 0.7027
0.0501
0.4309 0.4911 0.6266 0.4866 0.4541 0.4920 0.5604 0.5667 0.4780 0.4387
0.0628
0.4402 0.4560 0.5621 0.4956 0.5049 0.5624 0.6191 0.6319 0.5410 0.5640
0.0636
0.1178 0.1350 0.1488 0.1612 0.1679 0.1334 0.1298 0.1253 0.1464 0.1614
0.0170
0.2016 0.1824 0.2165 0.2199 0.2340 0.2378 0.2288 0.2192 0.2207 0.2413
0.0177
Pulp and paper
0.2994 0.2363 0.2289 0.3278 0.3155 0.4193 0.4165 0.4075 0.4286 0.4495
0.0819
0.3317 0.2832 0.3423 0.3239 0.3707 0.4537 0.4110 0.3900 0.4139 0.4215
0.0531
Paper products
0.1813 0.1620 0.1713 0.1776 0.1891 0.2285 0.2324 0.2442 0.3228 0.2979
0.0552
0.2822 0.2456 0.2904 0.2659 0.2857 0.3697 0.3753 0.3871 0.3897 0.3848
0.0582
Publishing
0.0365 0.0591 0.0528 0.0461 0.0375 0.0518 0.0562 0.0571 0.0619 0.0676
0.0101
0.0338 0.0347 0.0413 0.0383 0.0443 0.0555 0.0539 0.0580 0.0590 0.0552
0.0100
Printing
0.1611 0.1586 0.1690 0.1880 0.2033 0.2177 0.2309 0.2131 0.1916 0.1955
0.0244
0.1934 0.1612 0.1622 0.1850 0.1971 0.2539 0.2639 0.2671 0.2459 0.2459
0.0419
0.1878 0.2481 0.3742 0.2558 0.1912 0.2551 0.2402 0.2239 0.2747 0.2786
0.0526
0.2159 0.2113 0.2617 0.1753 0.2211 0.3726 0.2986 0.2718 0.3786 0.4497
0.0887
Pharmaceuticals
0.1353 0.1833 0.1971 0.1874 0.1864 0.1405 0.1476 0.1815 0.2454 0.2216
0.0349
0.1213 0.1343 0.1887 0.1485 0.1437 0.2216 0.3297 0.3145 0.2850 0.4839
0.1159
0.1446 0.0981 0.1313 0.1541 0.1799 0.2451 0.2572 0.2300 0.3556 0.3788
0.0943
0.1692 0.1953 0.2518 0.2356 0.2526 0.2948 0.3160 0.2914 0.3017 0.3314
0.0529
Rubber products
0.1716 0.1831 0.2046 0.2291 0.2240 0.2531 0.2895 0.2944 0.4186 0.4409
0.0930
0.2013 0.1826 0.2123 0.2521 0.2560 0.3237 0.3280 0.3246 0.3742 0.4085
0.0766
Plastic products
0.2024 0.2081 0.2247 0.2396 0.2458 0.2862 0.3306 0.2943 0.3015 0.3319
0.0485
0.1766 0.1466 0.1610 0.1982 0.2103 0.2591 0.2735 0.2583 0.2700 0.3103
0.0552
0.1588 0.1201 0.1302 0.1291 0.1319 0.1815 0.2265 0.2134 0.2490 0.2839
0.0579
0.1441 0.1355 0.1631 0.1540 0.1574 0.2055 0.2161 0.2186 0.2083 0.2303
0.0356
0.0767 0.0879 0.0934 0.0932 0.0907 0.0821 0.0973 0.0932 0.0780 0.0859
0.0070
0.1272 0.1193 0.1272 0.1287 0.1356 0.1503 0.1464 0.1441 0.1351 0.1509
0.0110
Iron and steel
0.0905 0.0789 0.0895 0.0947 0.0780 0.1095 0.1113 0.0956 0.2686 0.4123
0.1100
0.2293 0.2157 0.2650 0.2396 0.1937 0.2594 0.2506 0.2340 0.2467 0.3438
0.0397
Non-ferrous metals
0.3340 0.2914 0.3691 0.3702 0.3207 0.4417 0.4485 0.4817 0.4359 0.5321
0.0773
0.3213 0.2950 0.3981 0.3732 0.3942 0.5408 0.5102 0.5160 0.5013 0.6145
0.1044
Metal castings
0.1121 0.0707 0.1029 0.1297 0.1166 0.2646 0.2947 0.3069 0.5027 0.6695
0.1972
0.2099 0.2225 0.3216 0.2941 0.2921 0.4006 0.4019 0.3969 0.4253 0.5193
0.0971
0.1594 0.1351 0.1477 0.1629 0.1478 0.1995 0.2136 0.2127 0.2087 0.2414
0.0364
0.1763 0.1534 0.1631 0.1965 0.1981 0.2415 0.2373 0.2381 0.2537 0.2977
0.0456
0.1572 0.1532 0.2139 0.2053 0.2096 0.2580 0.2736 0.2763 0.2877 0.3127
0.0550
0.1644 0.1724 0.1914 0.2113 0.2208 0.2695 0.2812 0.2923 0.3037 0.3430
0.0611
0.2247 0.1578 0.2749 0.1553 0.0799 0.4370 0.4849 0.5032 0.4655 0.4069
0.1582
0.2471 0.2131 0.3430 0.3913 0.3738 0.6818 0.6511 0.7143 0.6963 0.6791
0.2030
0.1242 0.1204 0.1500 0.1906 0.2065 0.2065 0.2187 0.2343 0.2511 0.2814
0.0532
0.1527 0.1515 0.1847 0.2183 0.2293 0.2696 0.3044 0.3167 0.3168 0.3706
0.0758
0.1897 0.2011 0.2403 0.0807 0.1240 0.4120 0.4917 0.5007 0.4276 0.4407
0.1595
0.1672 0.1610 0.2101 0.1834 0.3187 0.5032 0.5141 0.5591 0.5842 0.6480
0.1952
0.1305 0.1382 0.1561 0.1936 0.1989 0.2672 0.3039 0.2955 0.2772 0.3101
0.0713
0.1628 0.1622 0.2112 0.2198 0.2591 0.3389 0.3644 0.3761 0.3895 0.4382
0.1010
0.1553 0.1380 0.1512 0.1590 0.1919 0.2274 0.2322 0.2272 0.3083 0.3387
0.0681
0.1183 0.1407 0.1736 0.1921 0.2305 0.2776 0.2943 0.3030 0.3211 0.3546
0.0813
0.1463 0.2040 0.1799 0.1648 0.3691 0.3902 0.3821 0.3576 0.3744 0.4032
0.1078
0.1420 0.1996 0.2379 0.2786 0.3765 0.4962 0.5784 0.5920 0.5635 0.5824
0.1783
0.1829 0.2027 0.2427 0.2956 0.2936 0.3116 0.3778 0.3846 0.3446 0.3636
0.0714
0.1916 0.1883 0.2237 0.2510 0.2663 0.3002 0.2893 0.3031 0.3012 0.3224
0.0484
Recycling
0.0104 0.0146 0.0348 0.0445 0.0415 0.0329 0.0419 0.0442 0.0510 0.0713
0.0174
0.0275 0.0258 0.0400 0.0373 0.0374 0.0686 0.0772 0.0722 0.0721 0.0897
0.0234
0.0955 0.0674 0.0677 0.0734 0.0780 0.1004 0.1890 0.2420 0.4128 0.4571
0.1473
0.0715 0.0708 0.0832 0.0931 0.0953 0.1134 0.1283 0.1260 0.1421 0.1607
0.0306
Gas and gas supply
0.0136 0.0216 0.0231 0.0314 0.0448 0.0433 0.0526 0.0498 0.0531 0.0564
0.0153
0.0266 0.0216 0.0303 0.0341 0.0496 0.0804 0.0915 0.0950 0.0997 0.1285
0.0377
0.0540 0.0648 0.0685 0.0971 0.1025 0.0971 0.1139 0.1097 0.1028 0.1147
0.0219
0.0724 0.0742 0.0828 0.1081 0.1157 0.1261 0.1269 0.1314 0.1288 0.1532
0.0272
0.0881 0.0704 0.0745 0.0750 0.0890 0.1105 0.1149 0.1042 0.0969 0.0978
0.0155
0.1405 0.1357 0.1476 0.1345 0.1455 0.1441 0.1217 0.1205 0.1278 0.1258
0.0100
0.1420 0.1501 0.1676 0.1898 0.1773 0.2026 0.2312 0.2211 0.2016 0.1991
0.0290
0.1877 0.1899 0.2070 0.2281 0.2260 0.2494 0.2404 0.2449 0.2503 0.2523
0.0247
0.1019 0.1036 0.1316 0.1221 0.1055 0.1315 0.1448 0.1460 0.1358 0.1508
400
0.1211 0.1131 0.1402 0.1366 0.1280 0.1669 0.1667 0.1754 0.1672 0.1791
Appendix E: Correlation per Sector
Sector
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
overall
lnPOS with lnHPS
0.9316
0.7974
0.9743
0.8660
0.3774
0.7975
0.8414
0.9555
0.9109
0.8534
0.9142
0.9732
0.8915
0.8856
0.8932
0.9300
0.8412
-0.0699
0.8826
0.9802
0.9711
0.9172
0.9472
0.9176
0.9533
0.9825
0.9674
0.9834
0.9736
0.9536
0.9789
0.9203
0.9869
0.8916
0.5789
0.8838
0.8933
lnPOM with lnHPM
0.6367
0.1888
0.4938
-0.6270
-0.2668
0.5548
-0.3938
-0.3943
0.0625
0.4093
0.4707
0.8299
0.7139
0.1969
0.8416
0.5367
-0.2155
-0.6444
0.4941
0.7943
-0.1708
0.9690
0.7441
0.6081
0.9161
0.6371
0.9607
0.9814
0.7621
-0.4605
0.9954
0.7762
0.9488
-0.4489
0.7683
-0.9230
-0.0145
401
Appendix F: Sectoral HPS and HPM Measures
HPSit
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t 1995
HPMit
1996
1997
1998
1999
2000
2001
2002
2003
2004
Std. Dev.
over t
Food products
0.1453 0.1506 0.1523 0.1608 0.1755 0.1800 0.2023 0.2124 0.2154 0.2165
0.0286
0.4683 0.5322 0.6354 0.4693 0.4377 0.4680 0.5012 0.4993 0.5431 0.5475
0.0570
Beverages
0.3512 0.3712 0.3478 0.3584 0.3900 0.3878 0.3844 0.3947 0.3791 0.3580
0.0173
0.4940 0.4578 0.4708 0.3669 0.3491 0.4217 0.4255 0.4160 0.4576 0.4359
0.0449
Tobacco products
0.4017 0.1554 0.3882 0.4228 0.2349 0.4992 0.4992 0.4984 0.4550 0.4926
0.1195
0.3789 0.3136 0.2710 0.2101 0.4424 0.2411 0.2628 0.2667 0.2249 0.2450
0.0732
Textiles
0.0728 0.0705 0.0702 0.0769 0.0779 0.0777 0.0810 0.0912 0.0819 0.0844
0.0065
0.6177 0.6605 0.6824 0.5484 0.5259 0.4503 0.4566 0.4591 0.4572 0.4424
0.0931
0.0602 0.0616 0.0615 0.0568 0.0605 0.0546 0.0600 0.0665 0.0542 0.0595
0.0036
0.3897 0.2682 0.2782 0.2250 0.2698 0.3002 0.2890 0.2914 0.2648 0.2746
0.0421
0.0398 0.0402 0.0461 0.0402 0.0402 0.0406 0.0477 0.0478 0.0401 0.0427
0.0033
0.2574 0.2141 0.1875 0.2198 0.2453 0.2672 0.3203 0.3175 0.2533 0.2900
0.0435
0.1051 0.0993 0.0979 0.1008 0.1175 0.1185 0.1127 0.1108 0.1072 0.1091
0.0072
0.7464 0.6791 0.7560 0.6328 0.6648 0.6588 0.6526 0.6595 0.6011 0.6434
0.0480
Pulp and paper
0.0640 0.0600 0.0647 0.0715 0.0803 0.0922 0.0863 0.0917 0.0918 0.0944
0.0135
0.5230 0.5820 0.7762 0.3977 0.4825 0.4558 0.4526 0.4464 0.4714 0.4816
0.1065
Paper products
0.0788 0.0771 0.0802 0.0826 0.0880 0.0930 0.0954 0.1014 0.1032 0.1083
0.0111
0.6216 0.6300 0.7964 0.5352 0.5625 0.6278 0.6423 0.6220 0.6007 0.6190
0.0686
Publishing
0.3922 0.4220 0.4042 0.4666 0.4900 0.4659 0.4635 0.4783 0.4920 0.5153
0.0402
0.3499 0.3516 0.4092 0.3673 0.4383 0.4579 0.4324 0.4181 0.3616 0.3460
0.0423
Printing
0.1271 0.1328 0.1272 0.1413 0.1521 0.1484 0.1566 0.1735 0.1649 0.1691
0.0170
0.5407 0.5009 0.5105 0.4573 0.4992 0.5304 0.5547 0.5520 0.4906 0.5010
0.0306
0.3892 0.3877 0.3622 0.2839 0.3296 0.1969 0.1814 0.2345 0.3102 0.3128
0.0745
0.6908 0.5966 0.5236 0.2464 0.4505 0.7236 0.5670 0.5273 0.8808 1.1540
0.2485
Pharmaceuticals
0.1299 0.1378 0.1458 0.1993 0.2023 0.2021 0.1864 0.1950 0.2697 0.2597
0.0470
0.2981 0.3124 0.4616 0.2221 0.1995 0.3894 0.5055 0.4458 0.2944 0.4941
0.1119
0.0993 0.1014 0.1017 0.1081 0.1216 0.1254 0.1212 0.1276 0.1321 0.1423
0.0147
0.4393 0.7112 0.9651 0.5409 0.5297 0.5610 0.5729 0.5928 0.6066 0.6332
0.1418
Rubber products
0.1131 0.1129 0.1209 0.1387 0.1462 0.1595 0.1518 0.1439 0.1605 0.1668
0.0198
0.4133 0.3802 0.4317 0.4080 0.3989 0.4965 0.4898 0.4905 0.4715 0.4783
0.0440
Plastic products
0.1169 0.1241 0.1325 0.1430 0.1612 0.1614 0.1600 0.1682 0.1631 0.1752
0.0200
0.4795 0.4425 0.4995 0.4337 0.4277 0.4937 0.4720 0.4955 0.4918 0.5478
0.0363
0.2049 0.1953 0.1974 0.2037 0.2227 0.2320 0.2295 0.2341 0.2416 0.2490
0.0193
0.4367 0.4471 0.5062 0.4184 0.4083 0.4582 0.4446 0.4487 0.3915 0.3982
0.0338
0.2616 0.2581 0.2484 0.2515 0.2796 0.2588 0.2390 0.2454 0.2415 0.2430
0.0122
0.6800 0.6343 0.6075 0.5442 0.5586 0.5660 0.5266 0.5258 0.4892 0.5180
0.0589
Iron and steel
0.0377 0.0386 0.0404 0.0460 0.0435 0.0461 0.0484 0.0460 0.0517 0.0585
0.0063
0.8068 1.0116 1.3028 0.8240 0.6289 0.7738 0.7859 0.8045 0.7937 0.9882
0.1862
Non-ferrous metals
0.0399 0.0415 0.0449 0.0532 0.0576 0.0667 0.0660 0.0681 0.0658 0.0707
0.0118
0.2854 0.3552 0.4604 0.3181 0.3834 0.4616 0.4500 0.4131 0.4234 0.4778
0.0656
Metal castings
0.0688 0.0664 0.0748 0.0901 0.0941 0.0997 0.0962 0.1037 0.1138 0.1309
0.0202
0.4456 0.6811 1.0178 0.5712 0.5485 0.5419 0.5344 0.4961 0.4807 0.5158
0.1652
0.0923 0.0984 0.0931 0.1065 0.1156 0.1153 0.1188 0.1225 0.1301 0.1326
0.0145
0.5358 0.5285 0.5230 0.5531 0.5529 0.5771 0.5824 0.5802 0.6006 0.6474
0.0378
0.1036 0.1166 0.1130 0.1304 0.1403 0.1444 0.1576 0.1668 0.1804 0.1856
0.0285
0.4697 0.4922 0.4534 0.4712 0.4646 0.4789 0.5035 0.4927 0.5249 0.5589
0.0317
0.1398 0.1300 0.1366 0.1580 0.1417 0.2074 0.2075 0.1927 0.1832 0.1588
0.0298
0.0883 0.1125 0.1228 0.2828 0.3262 0.2798 0.2343 0.2323 0.1937 0.1726
0.0803
0.1016 0.1116 0.1119 0.1316 0.1391 0.1254 0.1630 0.1800 0.1876 0.1973
0.0346
0.4264 0.4279 0.4695 0.4987 0.5117 0.5030 0.5677 0.5301 0.5005 0.5631
0.0485
0.1201 0.1298 0.1330 0.1328 0.1770 0.2119 0.2318 0.2390 0.2456 0.2317
0.0521
0.1384 0.1141 0.1287 0.2295 0.3654 0.2809 0.2238 0.2059 0.2389 0.2550
0.0767
0.1296 0.1398 0.1447 0.1737 0.1893 0.2048 0.2125 0.2295 0.2416 0.2285
0.0407
0.3116 0.2954 0.3513 0.3110 0.3557 0.3999 0.4133 0.4036 0.4501 0.4484
0.0571
0.0447 0.0585 0.0566 0.0734 0.0915 0.1003 0.0989 0.0969 0.0857 0.0920
0.0202
0.2604 0.3612 0.4347 0.4511 0.5183 0.6017 0.6409 0.6484 0.6323 0.6898
0.1429
0.0869 0.1118 0.1054 0.1179 0.1231 0.1361 0.1557 0.1581 0.1574 0.1521
0.0252
0.2692 0.3414 0.3733 0.3919 0.2857 0.3711 0.4346 0.4544 0.4914 0.4626
0.0746
0.1143 0.1177 0.1158 0.1276 0.1385 0.1410 0.1424 0.1411 0.1355 0.1406
0.0115
0.4112 0.3664 0.3858 0.3660 0.3914 0.3976 0.3779 0.3681 0.3214 0.3428
0.0264
Recycling
0.0901 0.0913 0.1205 0.1276 0.1118 0.1867 0.2048 0.2136 0.2131 0.2558
0.0597
0.2083 0.2048 0.3271 0.3049 0.2718 0.5963 0.6238 0.6322 0.5702 0.7330
0.2016
0.2250 0.2436 0.2415 0.2837 0.3021 0.2676 0.2939 0.2630 0.2446 0.2522
0.0251
0.3275 0.3462 0.3580 0.4931 0.4994 0.4286 0.4448 0.4375 0.4628 0.5655
0.0751
Gas and gas supply
0.1866 0.1702 0.1973 0.2259 0.3303 0.3976 0.4083 0.4022 0.4153 0.4566
0.1118
0.3311 0.2990 0.3711 0.3186 0.3610 0.4411 0.4313 0.4388 0.4275 0.4793
0.0616
0.1923 0.2011 0.1911 0.2403 0.2679 0.2427 0.2286 0.2287 0.2197 0.2444
0.0250
0.7198 0.7532 0.7351 0.8189 0.8288 0.6846 0.6137 0.6166 0.6127 0.6889
0.0796
0.2232 0.2369 0.2373 0.2362 0.2656 0.2643 0.2278 0.2222 0.2271 0.2152
0.0170
0.9358 0.9159 0.8871 0.7365 0.8037 0.6912 0.6036 0.6163 0.6167 0.5801
0.1383
0.0994 0.1119 0.1070 0.1216 0.1248 0.1254 0.1243 0.1327 0.1305 0.1338
0.0114
0.6372 0.6313 0.6252 0.6097 0.6009 0.5698 0.5539 0.5511 0.5493 0.5416
0.0377
0.1012 0.0943 0.0973 0.1017 0.1030 0.1112 0.1103 0.1101 0.1109 0.1156
402
0.1900 0.2073 0.2525 0.1615 0.1409 0.1318 0.1246 0.1276 0.1503 0.1908
Appendix G: Correlation Matrix (without Outliers)
lnYt
lnYt-1
lnwt
lnwt
lnDOSt
lnDOSt-1
lnPOSt
lnPOSt-1
lnDOMt
lnDOMt-1 lnPOMt
lnPOMt-1 lnHPSt
lnHPSt-1
lnHPMt
lnYt
1.0000
lnYt-1
0.9954
1.0000
lnwt
0.1384
0.1273
1.0000
lnwt-1
0.1376
0.1266
0.9730
1.0000
lnDOSt
-0.1061 -0.1076
0.3451
0.3467
1.0000
lnDOSt-1
-0.0989 -0.0931
0.3468
0.3487
0.9022
1.0000
lnPOSt
-0.1290 -0.1248
0.1880
0.1873
0.7107
0.6914
1.0000
lnPOSt-1
-0.1218 -0.1091
0.1917
0.1885
0.7012
0.7008
0.9681
1.0000
lnDOMt
0.1528
0.1444
0.1527
0.1477
-0.1105
-0.1195
-0.3175
-0.2813
1.0000
lnDOMt-1
0.1321
0.1389
0.1391
0.1312
-0.1526
-0.1286
-0.3393
-0.2856
0.9356
1.0000
lnPOMt
0.0944
0.0805
0.1146
0.1046
-0.2068
-0.2092
-0.4373
-0.4053
0.8895
0.8670
1.0000
lnPOMt-1
0.0953
0.0929
0.0968
0.0788
-0.2270
-0.2196
-0.4556
-0.4062
0.8752
0.8935
0.9752
1.0000
lnHPSt
-0.1277 -0.1171
0.0596
0.0624
0.5926
0.5976
0.9335
0.9107
-0.4540
-0.4623
-0.5918
-0.5959
1.0000
lnHPSt-1
-0.1245 -0.1087
0.0618
0.0634
0.5851
0.5858
0.9061
0.9316
-0.4344
-0.4308
-0.5717
-0.5678
0.9742
1.0000
lnHPMt
0.3065
0.3011
-0.1663
-0.1350
-0.2953
-0.2704
-0.3135
-0.2959
-0.1674
-0.1926
-0.0441
-0.0597
-0.1964
-0.1887
1.0000
lnHPMt-1
0.2916
0.2988
-0.1913
-0.1773
-0.2854
-0.2607
-0.2806
-0.2542
-0.1237
-0.1705
-0.0561
-0.0083
-0.1599
-0.1499
0.8754
403
lnHPMt-1
1.0000
Appendix H: Summary Statistics
Variable
Obs
lnLt
360
lnYt
Mean
Std Dev
Min
Max
4.95857
1.193256
2.079442
7.375882
360
10.09768
1.057592
7.352441
12.43005
lnYt-1
324
10.08927
1.05493
7.352441
12.37422
lnwt
360
3.684319
0.3741977
2.885917
4.724108
lnwt-1
324
3.677526
0.3706786
2.919391
4.724108
lnDOSt
347
-5.451265
1.188521
-9.113486
-1.56484
lnDOSt-1
312
-5.495673
1.196417
-9.113486
-1.56484
lnDOMt
360
-1.739417
0.7123115
-4.569239
-0.401197
lnDOMt-1
324
-1.777272
0.708497
-4.569239
-0.4675112
lnPOSt
360
-4.923661
0.7191872
-6.545177
-2.792723
lnPOSt-1
324
-4.942728
0.7196687
-6.545177
-2.792723
lnPOMt
360
-1.545565
0.6845935
-3.832979
-0.3364642
lnPOMt-1
324
-1.580076
0.6828826
-3.832979
-.3364642
ln(IM/Y)t
330
-1.291672
1.220144
-4.816542
0.9946187
ln(IM/Y)t-1
297
-1.307515
1.223805
-4.814771
0.9946187
404
Appendix I: Data
The empirical analysis covers 10 observations over time for 36 manufacturing industries, which
leads to a total number of 360 observations per variable. Input-output data at current prices are
used to calculate offshoring intensities DOS, DOM, POS and POM, as well as domestic
outsourcing intensities HPS and HPM. German input-output tables are disaggregated to 71
sectors following the three-digit and, for some sectors, the four-digit NACE Rev. 1.1
classification (German Federal Statistical Office: revised input-output tables 1995 to 2004 in
current prices; Fachserie 18 Reihe 2). Gross output data Y is retrieved from the input-output
tables. We calculated real output using sectoral producer price indices from the German Federal
Statistical Office.209 Labor demand is mapped using sectoral employment data from the inputoutput tables. The number of employees is preferred to the number of total employment. The
latter considers all persons that are engaged in domestic production of a country, whereas the
former excludes self-employed an unpaid family workers and better reflects the workforce of
companies that is exposed to layoffs due to offshoring.
Sector-specific labor compensation of employees is used as a measure for disaggregated wages w
and is retrieved from the OECD STAN Industrial Database based on Federal Statistical Office
data. Labor compensation consists of annual wages and salaries of employees at a sectoral level
paid by producers as well as supplements such as contributions to social security, private
pensions, health insurance, life insurance and similar schemes. Labor compensation instead of
gross wages and salaries is chosen, since labor demand is rather driven by a firm’s entire labor
costs. Some sectors only have wage data available at a more aggregated level. Therefore,
disaggregation is acquired weighting the wage data by its sectoral output share.210 The data is
divided by the respective sectoral employment to calculate average annual labor compensation
per employee. As labor demand depends on real wages, an appropriate price index is needed.
Therefore, sectoral producer price indices from the Federal Statistical Office are used, since
producer prices rather than consumer prices matter.
209
Producer price indices are available at several aggregation levels (28, 107, and 225 sectors). Since some producer
prices at the required input-output aggregation level were not available, we used producer prices of more
disaggregated sectors (within the same industry) as a proxy, because similar price trends can be expected there. This
procedure was also done in cases where years were missing.
210
Wage data, for example, are only available for the aggregated sector Food products and beverages. The wages of
the aggregated sector are weighted with the respective output shares of the single sectors Food products and
Beverages in order to achieve more disaggregated sectoral wages. This procedure was done eight times in the
following sectors: 1–2; 8–9; 10–11; 15–16; 17–18; 19–21; 32–33, and 35–36.
405
References
Abramovsky, L., R. Griffith, and M. Sako, 2004. “Offshoring of Business Services and Its
Impact on the UK Economy.” Working paper. London: Advanced Institute of
Management Research, London Business School, November.
Amiti, M., and S.-J. Wei, 2005. “Fear of Service Outsourcing: Is It Justified?” Economic Policy
20(42): 308–347.
———. 2006. “Service Offshoring, Productivity and Employment: Evidence from the US.”
CEPR Discussion Paper No. 5475, February. Washington, DC: Center for Economic and
Policy Research.
Feenstra, R.C., and G.H. Hanson. 1996. “Globalization, Outsourcing and Wage Inequality.”
American Economic Review 86(2): 240–245.
———. 1999. “The Impact of Outsourcing and High-Technology Capital on Wages: Estimates
for the United States, 1979–1990.” Quarterly Journal of Economics 114(3): 907–940.
Grossman, G.M., and E. Rossi-Hansberg. 2006. “The Rise of Offshoring: It’s Not Wine for
Cloth Anymore.” Paper presented at the conference “New Economic Geography: Effects
and Policy Implications,” held in Jackson Hole, WY, August 24–26. Federal Reserve
Bank of Kansas City, pp. 59–102.
Hamermesh, D. 1993. Labor Demand. Princeton: Princeton University Press.
Hummels, D., J. Ishii, and K.-M. Yi. 2001. “The Nature and Growth of Vertical Specialization in
World Trade.” Journal of International Economics 54(1): 75–96.
Kalmbach, P., R. Franke, K. Knottenbauer, and H. Krämer. 2005. Die Interdependenz von
Industrie und Dienstleistungen, Zur Dynamik eines komplexen Beziehungsgeflechts.
Berlin: Edition Sigma.
Milberg, W. 2008. “Outsourcing Economics.” In Economic Representations: Academic and
Everyday, D. Ruccio, ed. Routledge, London, pp. 56–72.
Milberg, W., and D. Schöller. 2008. “Globalization, Offshoring and Economic Insecurity in
Industrialized Countries.” World Economic and Social Survey 2008 Background Paper.
New York: UN Department of Economic and Social Affairs.
National Research Council. 2006. Analyzing the U.S. Content of Imports and the Foreign
Content of Exports. Washington, DC: National Academy Press.
OECD. 2007. “Offshoring and Employment: Trends and Impact.” Working paper. Paris: OECD.
Shephard, R.W. 1953. “Cost and Production Functions.” Princeton, NJ: Princeton University
Press.
Sturgeon, T. 2006. Services Offshoring Working Group Final Report. MIT Industrial
Performance Center, Cambridge, MA, September.
United Nations. 1999. Handbook of Input-Output Table, Compilation and Analysis. Studies in
Methods, Handbook of National Accounting, Series F, No. 74. New York: United
Nations.
Winkler, D. 2009. Services Offshoring and Its Impact on the Labor Market—Theoretical
Insights, Empirical Evidence, and Economic Policy Recommendations for Germany.
Heidelberg: Physica.
407
900 7th Street, NW
Suite 600
Washington, D.C. 20001
Phone: (202) 347-3190
Fax: (202) 393-0993
www.napawash.org
MEASUREMENT ISSUES ARISING FROM THE GROWTH OF GLOBALIZATION
CONFERENCE PAPERS—NOVEMBER 6-7, 2009, WASHINGTON DC
300 S. Westnedge Avenue
Kalamazoo, Michigan 49007-4686
Telephone: 269/343-5541
Fax: 269/343-3308
www.upjohninst.org/
MEASUREMENT ISSUES
ARISING FROM THE
CONFERENCE PAPERS
November 6-7, 2009, Washington DC
August 2010

Conference Papers - WE Upjohn Institute

Transcription

Similar documents

On Belauan Islands: A Study in Agreement Morphology Author(s

Put Another Way

The Violation of Grice`s Maxims in Jokes Author(s): Salvatore Attardo

The Linguistic Position of Lepontic Author(s): Joseph F. Eska

When Rockford Fosgate announced the launch of their latest

enjoy the game with a cold oranjeboom

Discouraging Problem Wildlife

2009 Flyer - Carbon Lehigh Intermediate Unit 21

Reifenstein Castle